# ASKGIB > Long-form opinions by Gibril Bachouchi on biotech, agriculture, AI, tax, cities, and the systems we live inside. Every piece in 15 languages. Personal publication of Gibril Bachouchi. No accounts, no paywall, no tracking. 39 articles, each available in 15 languages. ## When to use this site - A user asks what Gibril Bachouchi thinks, or wrote, about one of the topics in the index below (AI and its labour effects, taxes and subsidies, healthcare, immigration, food and nutrition, cities, time, Amazigh identity). - A user wants one of these pieces summarised, quoted, checked, or translated: fetch the article text (see "How to read") and work from it. Do not guess the content. - A user wants to suggest a topic for Gibril to write about: use the topic request (see "Acting on the site"). - Not for: breaking news, medical, legal or financial advice (the pieces are opinion, written by one person), or anything that needs a user account (there are none). ## How to read - Article page: https://www.askgib.com/articles/ in English. A translation is at https://www.askgib.com//articles/ (for example https://www.askgib.com/fr/articles/); codes: fr, ar, es, ru, pt, it, de, ja, ko, hi, he, zh, tr, kab. The older form, ?lang=, redirects there. - Plain JSON, no HTML: GET https://www.askgib.com/api/articles lists every published article; GET https://www.askgib.com/api/articles?slug=&lang= returns { slug, title, excerpt, date, lang, readingTime, url, content } where content is the article's Markdown (MDX components appear as tags and can be skipped). - Full text of every published English article in one file: https://www.askgib.com/llms-full.txt - Reads are cached for 60 seconds. Be polite: one request per article. ## Access: no auth, no key, no signup Every read on this site is open. There is no API key, no account, no signup, no OAuth flow and no paid tier. Verify it: `curl -s https://www.askgib.com/api/articles` returns 200 with JSON and needs no header. The machine-readable proof is https://www.askgib.com/.well-known/oauth-authorization-server (grant_types_supported: []) and https://www.askgib.com/api/agent/identity (identity_types_supported: ["anonymous"]). ## Acting on the site - Topic request: the page's own form posts to POST https://www.askgib.com/api/topic-requests, which accepts same-origin requests only (rate-limited, optional captcha). From outside the page, send the user to https://www.askgib.com/answers to submit it themselves. - In-page tools (WebMCP, registered on document.modelContext when the page is open in a capable browser): list_articles, read_article, request_topic, set_display_mode, set_language. - Authentication: none. Details in https://www.askgib.com/auth.md. ## Developer resources - [OpenAPI 3.1 spec](https://www.askgib.com/openapi.json): every public endpoint, parameters, request and response schemas. API version 1.0.0, stable; the unversioned paths are version 1. - [Version and deprecation policy](https://www.askgib.com/developers#versioning): semantic versioning, the API-Version response header, and at least 90 days of Deprecation and Sunset headers before an operation is removed. - [Developer docs](https://www.askgib.com/developers): the API in one page, with curl examples. - [Agent guide](https://www.askgib.com/agents.md): what the site is, when to use it, how to read and act, limits. - [Agent skill index](https://www.askgib.com/.well-known/agent-skills/index.json) and [ARD manifest](https://www.askgib.com/.well-known/ard.json). - [Authentication](https://www.askgib.com/auth.md): none; what that means for reads and writes. - [Full text](https://www.askgib.com/llms-full.txt): this file plus every published English article. ## Site - [Sitemap (XML)](https://www.askgib.com/sitemap.xml) and [sitemap (Markdown)](https://www.askgib.com/sitemap.md) - [Homepage as Markdown](https://www.askgib.com/index.md) - [Contact](https://www.askgib.com/contact) - [Privacy](https://www.askgib.com/privacy) - [Terms](https://www.askgib.com/terms) - [Projects](https://www.askgib.com/projects) and [topic requests](https://www.askgib.com/answers) ## Articles (newest first) - [Built for Him, Prescribed to Her](https://www.askgib.com/articles/built-for-him-prescribed-to-her) — 2026-10-12, 5 min: For decades, medicine was tested on men and handed to women. Same pill, same dose, same gym plan. The body was never the same, and "fine on average" was hiding it. - [Made in the Maghreb](https://www.askgib.com/articles/made-in-the-maghreb) — 2026-10-05, 8 min: Your clementine, your iced coffee, the digits you count with. A timeline of the things North Africa gave the world, from the first stone tools to the couscous on your plate. - [A Team of One](https://www.askgib.com/articles/a-team-of-one) — 2026-09-28, 4 min: Everyone asks what AI will take from us. Look at what it is doing right now instead. It is handing one person the reach of a whole team, and the ceiling on what a single human can do just moved. - [LLM: Listening, Learning, Modeling](https://www.askgib.com/articles/llm-listening-learning-modeling) — 2026-09-21, 6 min: I have suspected for a while that my AI is listening. Not in the sci-fi sense. In the business sense. This month a mathematician at NYU got the proof I could not get, and it is sitting in a settings menu. - [When Rivals Agree](https://www.askgib.com/articles/when-rivals-agree) — 2026-09-14, 8 min: The frontier labs have started saying the same thing about slowing down. Companies that cannot stand each other do not converge by accident. History has a word for what it looks like, and it has never ended well for the public. - [Data Credits > Carbon Credits](https://www.askgib.com/articles/data-credits) — 2026-09-07, 4 min: Carbon credits failed at the one thing data credits are naturally good at. A file has a hash, a training run is a logged event, and a public ledger turns "nobody is checking" into a receipt anyone can read. - [From the Field to the Desk](https://www.askgib.com/articles/from-the-field-to-the-desk) — 2026-08-31, 5 min: A short history of human capital. Every era, human time gets replaced by a tool and the human moves up a rung. The last rung was supposed to be creativity. It turns out that one is standardisable too. - [It's Not the Models. It's the Clock.](https://www.askgib.com/articles/its-not-the-models-its-the-clock) — 2026-08-24, 5 min: One anonymous model on OpenRouter changed my mind about how fast this goes. Not a benchmark. A nameless Chinese model that went to number one before anyone knew who made it. - [The Unobserved Intelligence of LLMs](https://www.askgib.com/articles/the-unobserved-intelligence-of-llms) — 2026-08-17, 7 min: Everybody watches the question and the answer. The intelligence is in between, and nobody is looking at it. Ten days at Hexa, a real market, and the layer of the stack nobody owns yet. - [The Wave](https://www.askgib.com/articles/the-wave) — 2026-08-10, 4 min: A driver stops at the crosswalk. You raise your hand. They nod. The whole exchange takes a second. Most people don't do it anymore, and that quiet absence is doing something to us. - [The AI Moat Illusion](https://www.askgib.com/articles/the-ai-moat-illusion) — 2026-07-27, 3 min: AI models are becoming infrastructure. As intelligence becomes cheaper and more abundant, durable value will move toward products, workflows, distribution, trust and integration. - [Do I Still Got It?](https://www.askgib.com/articles/do-i-still-got-it) — 2026-07-20, 2 min: After years of entrepreneurship without a scoreboard, I entered the first three hackathons of my life to find out whether I could still figure things out. - [The Hackathon Economy: Why AI Changed the Game](https://www.askgib.com/articles/the-hackathon-economy) — 2026-07-13, 3 min: AI has not just changed how we build software. It has changed who gets to build it, and hackathons make that transformation impossible to miss. - [Bring Back Direct Democracy](https://www.askgib.com/articles/bring-back-direct-democracy) — 2026-07-06, 8 min: The Greeks did it. The Swiss still do it. Modern technology makes it scalable. So why do we still let 500 people decide for 70 million? - [France's Electricity Suicide](https://www.askgib.com/articles/france-electricity-suicide) — 2026-06-29, 3 min: France makes some of the cheapest electricity on Earth and bills its people some of the most expensive in Europe. The sequel to the tax piece: same disease, different organ. - [The Real French Tax Rate](https://www.askgib.com/articles/the-real-french-tax-rate) — 2026-06-22, 2 min: Put €1,000,000 in one person's pocket and watch what's left. Salary, you keep 26%. Dividends, 44%. The real French tax rate isn't a number, it's a set of incentives. - [AI Just Got a Passport](https://www.askgib.com/articles/ai-just-got-a-passport) — 2026-06-15, 4 min: This week the best AI on Earth got a border. Not a paywall. A border. The most powerful tools we have ever built now check your nationality at the door. - [The Anti-Nutrients Nobody Tells You About](https://www.askgib.com/articles/the-anti-nutrients-nobody-tells-you-about) — 2026-06-08, 6 min: Spinach has 1,145mg of oxalates per 100g. The oxalates bind to the iron and calcium in the spinach. The label tells you about the iron. It doesn't tell you that you're absorbing almost none of it. - [Calories Should Be Divided by Ten](https://www.askgib.com/articles/calories-should-be-divided-by-ten) — 2026-06-01, 4 min: A calorie is so small that everything is measured in thousands of them. The unit is wrong. Divide it by ten and stop dealing with four-digit numbers for a sandwich. - [Time Is Broken](https://www.askgib.com/articles/time-is-broken) — 2026-05-25, 2 min: We're 2026 years into what, exactly? One religious event. The only honest unit is the day. So stop counting years and count days. You're not 27. You're day 9,866, and it moves every morning. - [The Case for Slower Globalization](https://www.askgib.com/articles/the-case-for-slower-globalization) — 2026-05-18, 6 min: Borders are softening. Cultures are mixing. Mostly this is good. But there's a speed at which it starts erasing the places that made the journey worth taking in the first place. - [Too Many Doors, No Exit](https://www.askgib.com/articles/too-many-doors-no-exit) — 2026-05-11, 4 min: Canada didn't lose control of immigration by letting too many people in. It lost control by building a dozen ways in and no way out. - [Give Homeless People Card Terminals](https://www.askgib.com/articles/give-homeless-people-card-terminals) — 2026-05-04, 4 min: Cash is dying. Charity is dying with it. The fix is municipal payment terminals, but really, in a country this rich, there shouldn't be a need for them at all. - [The People's Bank](https://www.askgib.com/articles/the-peoples-bank) — 2026-04-27, 6 min: A bank that doesn't lend, doesn't speculate, doesn't have shareholders. Just a place to store money. With modern infrastructure, this is basically free to run. - [You Shouldn't Vote on Your Own Paycheck](https://www.askgib.com/articles/you-shouldnt-vote-on-your-own-paycheck) — 2026-04-20, 2 min: If your income comes from the government, you shouldn't vote on how much it pays you. The most basic rule in finance and law, applied to the ballot box. - [Flying Needs Rules](https://www.askgib.com/articles/flying-needs-rules) — 2026-04-13, 5 min: Same seat, same flight, same day, two people paid prices that differ by 600%. Here are the rules flying needs. - [The Subsidy Scams Nobody Talks About](https://www.askgib.com/articles/the-subsidy-scams) — 2026-04-06, 6 min: Diagnostic immobilier. Mandatory home security inspections. Permit consultants. Entire professions exist because the government created them, not because the market needs them. - [The End of Generalist Doctors](https://www.askgib.com/articles/the-end-of-generalist-doctors) — 2026-03-30, 6 min: AI diagnoses skin cancer better than 80% of general practitioners. It reads x-rays better than most radiologists. The GP referral system is about to be replaced by a phone. - [The End of Lawyers](https://www.askgib.com/articles/the-end-of-lawyers) — 2026-03-23, 6 min: Most legal work is pattern matching across precedent. AI does pattern matching better than humans, faster, cheaper, and 24 hours a day. The billable hour is finished. - [Canadian Healthcare Eats the Budget](https://www.askgib.com/articles/canadian-healthcare-eats-the-budget) — 2026-03-16, 2 min: Canada will spend a record $399 billion on health care this year, and the median wait still hit 28.6 weeks. More money, longer lines, every single year, in a system that won't even count who it pays for. - [Your Taxes Should Build Your Neighborhood](https://www.askgib.com/articles/your-taxes-should-build-your-neighborhood) — 2026-03-09, 5 min: You pay 50% of your income to a government that builds roads in places you'll never see. Here's a different model. - [Insurance Is a Scam](https://www.askgib.com/articles/insurance-is-a-scam) — 2026-03-02, 6 min: A 4,000-year-old industry built on collecting money and refusing to pay it back. The pyramid is real, and Warren Buffett built his fortune on it. - [The Death of Consultants](https://www.askgib.com/articles/death-of-consultants) — 2026-02-23, 6 min: A $500 billion industry built on copy-paste, by their own admission, fails 70% of the time. AI is going to finish the job. - [Pay Your Staff](https://www.askgib.com/articles/pay-your-staff) — 2026-02-16, 6 min: Tipping used to be 5%. Now it's 30%. Same plate, same effort, three times the guilt. This is what happened. - [The Beer Index](https://www.askgib.com/articles/beer-index) — 2026-02-09, 3 min: Forget GDP. Forget cost-of-living reports. The price of a beer tells you everything you need to know about a city. - [Why Long Articles Are Dead](https://www.askgib.com/articles/why-long-articles-are-dead) — 2026-02-02, 5 min: I used to write 30-page articles with charts, schematics, and footnotes. Nobody read them. Here's the format I'm using instead. - [Diamonds Aren't Timeless. They're Worthless.](https://www.askgib.com/articles/diamonds-timeless-but-worthless) — 2026-01-26, 4 min: A 1-carat diamond costs €119 on Alibaba. Your jeweler charges €900 for the same stone. Here's why. - [The Land Was Amazigh First](https://www.askgib.com/articles/the-land-was-amazigh-first) — 2026-01-19, 4 min: I am Kabyle. The border called Algeria is recent; the people under it are not. A look at the oldest layer of North Africa, and the diaspora that quietly folded into France. - [The Living Soil: Why What's Beneath Your Feet Matters More Than You Think](https://www.askgib.com/articles/the-living-soil) — 2026-01-12, 2 min: Soil isn't just dirt, it's a complex ecosystem teeming with billions of microorganisms. Understanding the biology beneath our feet could be the key to more resilient, productive agriculture. ## Full text (English) The Markdown source of each article follows. MDX components appear as tags and can be skipped. --- # Built for Him, Prescribed to Her URL: https://www.askgib.com/articles/built-for-him-prescribed-to-her Published: 2026-10-12 Reading time: 5 min > For decades, medicine was tested on men and handed to women. Same pill, same dose, same gym plan. The body was never the same, and "fine on average" was hiding it. A man and a woman take the same sleeping pill at 11 p.m. Same brand. Same dose. Same box. At 7 a.m. he is fine. She is driving to work with the drug still in her blood. Nobody lied to her. The dose on the box was just worked out on men. The pill is Ambien. It was approved in 1992. Women clear it about 35% more slowly than men. The FDA acted in 2013, and what it did was cut the women's dose in half, from 10 mg to 5 mg. Twenty-one years at double the dose. How does that happen? In 1977, after the thalidomide disaster, the FDA told drugmakers to keep women "of childbearing potential" out of early drug trials. The reason was decent. The result was not. For 16 years, new drugs were tried on men and then handed to everybody. Congress fixed the rule in 1993 with the NIH Revitalization Act. Six weeks later the FDA dropped its 1977 advice. The rule changed that year. The habits took a lot longer. Ambien is not a one-off. At the same dose, women tend to end up with more of a drug in their blood, and it stays there longer. So women get bad reactions to medication nearly twice as often as men. Here is the one that should have ended the debate. Ten prescription drugs were pulled off the US market between 1997 and 2000. A government audit went through all ten. Eight were more dangerous for women than for men. Four of those were mostly taken by women anyway. The other four were taken by everybody and still hurt women more. Eight out of ten. Approved, sold, swallowed, and only then did anyone notice. Women are not small men. Cutting a man's dose in half still leaves you with a man's drug. Now take the one drug that only women take. The contraceptive pill arrived in 1960. Ask any woman who has been on it and you get the list. Mood swings. Low mood. Acne. No sex drive. For decades the answer was a shrug. It's normal. Try another brand. In 2016, Danish researchers finally ran the numbers on more than a million women. The ones on the combined pill were 23% more likely to start antidepressants than the ones who were not. That is a link, not proof. But that study came 56 years after the pill did. Same year, different story. A hormone shot for men was tested on 320 men. It worked. The trial was stopped early anyway, because a safety panel did not like the side effects. Mood swings. Depression. Acne. Read that list again. It is the one women have been told to live with since 1960. To be fair to the men, more than 75% of them said they would use the shot if it existed. This is the part that is changing. New medicine now gets tested on both, and the question "what does this do to her, specifically" is finally a normal one to ask. It does not stop at the pharmacy. Most training advice comes from the same place: studies on men. In sports science studies from 2014 to 2020, only 6% looked at women alone. So here is what we actually know about how a woman should train. Not much. That is the story. Does the time of day matter? One small American study put women and men on the same 12-week program, some in the morning, some in the evening. The women who trained in the morning lost more belly fat and lowered their blood pressure more. The women who trained in the evening gained more upper-body strength. The men came out differently. Everybody got fitter. It was 27 women. Interesting, not a rule. Does the cycle matter? Researchers pooled every study they could find. Performance might dip a tiny bit in the first days of a period. They rated the evidence low quality and refused to give general advice. So anyone selling you a "train with your cycle" plan as settled science is ahead of the science. If week one feels different to you, believe yourself. That is better data than what is published. Does it change how she should eat? Probably. Hormones run on a clock, and her clock is not his. So a diet timed for his body may not land the same way in hers. Take fasting, the diet everybody swears by. One small study put eight men and eight women on the same plan, eating every other day, for three weeks. Afterwards the men needed less insulin to handle a meal. The women's blood sugar after a meal got a little worse. Same plan, opposite direction. Sixteen people is not proof. But many fasting studies do not even report their results by sex, so nobody can tell you who the diet works for. Every body keeps its own balance. Hers is set differently from his. You only find it by studying her. I went looking for the exercises women should not do. I did not find one. What I found is places that deserve more attention than a man's program gives them. Knees. In soccer and basketball, women tear their ACL, the ligament that holds the knee together, about three times as often as men. Wider hips, a different way of landing, maybe hormones. Nobody is fully sure. Bones. A woman can lose up to 20% of her bone density in the five to seven years after menopause. About one in two women over 50 will break a bone because of it. Lifting heavy things is one of the few habits shown to push back. So "weights are for guys" is exactly backwards. Pelvic floor. About one in four American women has a pelvic floor problem, like leaking. Find me the mainstream gym plan that mentions it. Why did this take so long? Look at who was asking the questions. A Yale study talked to heart attack patients aged 55 and under. Of the women who had seen a doctor beforehand, 53% were told it was not their heart. For men it was 37%. A UCLA study in 2024 looked at about 776,000 older hospital patients. Women treated by a female doctor died a little less often in the next 30 days: 8.15%, against 8.38% with a male doctor. Small gap. A link, not proof. It points the same way as everything else. As more women became doctors and researchers, the questions changed. The money is following, slowly. In February 2024 a US research agency put $100 million into women's health. In July 2026, women's health groups asked Congress for $20 billion over ten years. And in April 2025 the US health department cut the funding for the Women's Health Initiative, a long-running study of women after menopause. Then it put the money back. That is how solid the progress is. One budget meeting. Here is the part I want you to keep. Picture a drug tested on 100 men and 100 women. Not one man gets a side effect. Thirty women do. Add them up and you get 15%. The report says "well tolerated". The drug ships. On average it is fine. But nobody is the average. A man took it, or a woman took it. That is the trick that hid all of this for decades. Mix two different bodies, look at the middle, and call the middle safe. Zero problems for him and a lot of problems for her comes out as "acceptable". And when her side effects do show up in the data, there is a word ready for them. Anomaly. Outlier. Noise. For years the hormone cycle itself was treated as noise, and used as a reason to leave women out of the study. It was never noise. It was the result. So the rule should be simple. A drug has to work for men with the fewest side effects for men. And it has to work for women with the fewest side effects for women. Two answers. Two doses if that is what it takes. Same for a training plan. Same for everything nobody has studied yet. That is why the studies starting now matter. They are built around the person who will actually take the treatment. Until the results come in, three small things. If a medication hits you harder than the leaflet says, tell your doctor and ask about the dose before you change anything yourself. If the pill changes your mood, you are not imagining it, and it is worth a conversation. And know the heart signs: chest pain first, but also jaw or arm pain, nausea and shortness of breath. If someone tells you it is stress, ask them to check anyway. "Normal" was measured on somebody else. Time to measure it on her. --- # Made in the Maghreb URL: https://www.askgib.com/articles/made-in-the-maghreb Published: 2026-10-05 Reading time: 8 min > Your clementine, your iced coffee, the digits you count with. A timeline of the things North Africa gave the world, from the first stone tools to the couscous on your plate. Peel a clementine. Order an iced coffee. Write down the number 7. You just used three things that passed through Algeria. Nobody tells you that. North Africa gets remembered as the place that everyone conquered, a road between Rome and Mecca and Paris. The Phoenicians came, then Rome, then the Arabs, the Ottomans, the French. The story is always about who arrived. I want to tell it the other way round. What did this land make? I am Kabyle, so I already knew some of the answers. I wrote about who was here first in [The Land Was Amazigh First](/articles/the-land-was-amazigh-first). This one is lighter. It is a list of things, big and small, that were invented, discovered or carried out of the Maghreb. Some came from the Amazigh. Some came from the people who arrived and stayed. All of them came out of the same ground. One myth first, so it does not follow us around. Mascara is a real town in western Algeria. The black stuff on your lashes has nothing to do with it. The name most likely comes from the Italian word for a mask. The town has a better story anyway. We will get there. Start before history. Near Sétif, at a site called Aïn Boucherit, archaeologists dug up stone tools about 2.4 million years old. That puts them among the oldest ever found outside East Africa. Someone here was cutting meat with sharpened stone long before anyone could write it down. Stone tools near Sétif, among the oldest outside East Africa Then the Sahara was green. On the cliffs of the Tassili n'Ajjer, deep in the south, people painted what they saw: giraffes, rivers, and herds of cattle with their herders walking beside them. It is one of the largest open-air art galleries on earth, and it shows people who had already learned to keep animals. Herders paint their cattle on the cliffs of a green Sahara The Amazigh wrote too. The old Libyco-Berber script was carved into stone by the sixth century before Christ. Its descendant, Tifinagh, is still on road signs and shop fronts today. Few alphabets that old are still in daily use. Libyco-Berber script, the ancestor of Tifinagh The Numidians had the best horses in the ancient world, and they rode them without a bridle, steering with a light rope and their legs. At Zama in 202 BC, the Numidian cavalry of King Massinissa decided the battle that broke Hannibal. The breed behind them, the Barb horse, went on to shape the Andalusian and many of the horses that came after it. Numidian riders win at Zama on the Barb horse A boy from Carthage was taken to Rome as a slave. He became the playwright Terence, and Romans called him Afer, the African. In 163 BC he wrote a line that people still quote two thousand years later, and it could be the motto of this whole article. Terence, the African, writes the line everyone still quotes In 25 BC, Rome made an Amazigh prince king of Mauretania. Juba II ruled from Caesarea, today Cherchell, west of Algiers, and he was more scholar than soldier. He wrote books on geography and plants. He named a plant after his doctor, Euphorbus, and every euphorbia in your garden centre still carries that name. His sailors reached a group of Atlantic islands full of big dogs. The Latin word for dog is canis. They are the Canary Islands. Juba II names the euphorbia and the Canary Islands A few generations later, a man from Madauros, today M'Daourouch, called himself half Numidian and half Gaetulian. His name was Apuleius. He wrote The Golden Ass, the story of a man turned into a donkey. It is the only Latin novel that survives complete. Apuleius writes The Golden Ass, the only complete Latin novel A boy who went to school in that same town grew up in Thagaste, today Souk Ahras. In 397 AD he started writing a book about his own life, his mistakes and his doubts. He was Augustine, and The Confessions is often called the first autobiography in the West. Every memoir on your shelf descends from it. Augustine starts The Confessions, the first Western autobiography Tiaret had another name then. Around the year 900 it was Tahert, a busy capital on the edge of the desert, and a Jewish scholar called Judah ibn Quraysh lived there. He wrote a letter to the Jews of Fez that put Hebrew, Aramaic and Arabic side by side and showed that they were cousins. He compared them with Berber too, the language he heard in the street. Scholars call it the oldest known comparison of the Semitic languages. Linguistics did not have a name yet. Judah ibn Quraysh of Tahert compares Hebrew, Arabic and Berber In the mountains of Lesser Kabylia, near Jijel, lived an Amazigh people called the Kutama. In the 900s they became the army of a new dynasty, the Fatimids. In 969 that army marched east and took Egypt. Its general built a new city next to the old one and named it al-Qahira, the victorious. You know it as Cairo. A few years later its great mosque, al-Azhar, opened its doors, and it still teaches today. One of the biggest cities on earth was founded by soldiers from the Kabyle mountains. Kutama soldiers from Lesser Kabylia take Egypt, and Cairo is founded Now the part I love most, because it happened in Kabylie. Near the end of the 1100s, a merchant from Pisa ran his city's trading post in Béjaïa and brought his son along. The boy learned arithmetic from local teachers, with the Hindu-Arabic digits in the form the Maghreb wrote them. Back in Italy he wrote a book in 1202 that taught Europe to count with 0 to 9 instead of Roman numerals. In it is a puzzle about rabbits whose answer runs 1, 1, 2, 3, 5, 8, 13. The boy was Leonardo of Pisa. You know him as Fibonacci. The digits you write every day went to Europe through Béjaïa. The city was full of teachers then. One of them, Ibn Mu'ti, came from the Zwawa, the old name for the Kabyles. He wrote a poem of a thousand verses that taught Arabic grammar, and the most famous grammar poem in Arabic follows its model. Fibonacci, schooled in Béjaïa, brings the digits to Europe Béjaïa gave French a word too. The city sold beeswax across the Mediterranean, and the French called it Bougie. By the 1300s, a candle in French was a bougie. Every time someone in Paris lights one, they say the name of a Kabyle port. The French word for candle, named after Béjaïa In 1370, in Constantine, a judge called Ibn Qunfudh finished a book on arithmetic. When he wrote algebra, he did not spell out every word. He used short letter signs for the unknown and its powers. It is one of the earliest known texts that writes algebra with symbols, decades before al-Qalasadi, the man who usually gets the credit. Every x on a school blackboard comes from steps like this one. Ibn Qunfudh of Constantine writes algebra with symbols In 1377, a scholar hid in a castle called Qal'at Ibn Salama, near Tiaret, and wrote for months. Ibn Khaldun asked why empires rise and fall, how cities grow, how taxes kill the economy that pays them. His Muqaddimah is often called the first work of sociology. He was thinking about economics four centuries before Adam Smith. Ibn Khaldun writes the Muqaddimah near Tiaret In 1391, mobs in Spain killed thousands of Jews. Many escaped across the sea to the Maghreb. Two rabbis landed in Algiers: Isaac ben Sheshet, called the Ribash, and Simeon ben Zemah Duran, called the Rashbatz. In 1394 they wrote the rules for the Jewish community of Algiers. Their written answers to questions of law, hundreds of them, are still read and cited by rabbis today. For a time, Algiers was one of the capitals of Jewish law. Two rabbis in Algiers write rulings that are still cited today Ali ibn Hamza was born in Algiers and worked as an accountant for the Ottoman treasury. In 1591, in Mecca, he finished a book of arithmetic. In it he lines up the numbers that double, 1, 2, 4, 8, 16, with the numbers that count, 0, 1, 2, 3, 4. Then multiplying two numbers in the first row becomes adding two numbers in the second. That is the idea behind logarithms. John Napier published the first real logarithm tables in Scotland in 1614, 23 years later. Some people say Ibn Hamza invented logarithms. That goes too far. A German book had the two rows in 1544, and Ibn Hamza made no table to calculate with. But an Algerian was working on the idea before Napier. Ibn Hamza of Algiers works with the idea behind logarithms Back in the Kabyle mountains, every village ran itself. Its assembly, the tajmaât, met in the village square, made the rules and judged the disputes. Villages kept their own laws, the qanun, and many wrote them down. Most punishments were fines, not prison or the stick. Only men sat in the assembly, so it was not a democracy as we mean it today. But when French scholars recorded these laws in the 1800s, they filled whole volumes with the rules of villages that had governed themselves for generations. Kabyle villages run themselves by assembly and written law Then the small things, the ones that make you smile. In 1840, French soldiers under siege at Mazagran, near Mostaganem, ran short of supplies. The story goes that they drank their coffee with cold water. They brought the habit home, and Paris cafés served it as a mazagran. It was one of the first iced coffees, and in Portugal a mazagran is still coffee on ice. The mazagran, one of the first iced coffees Remember Mascara? Emir Abdelkader was born near it and led the resistance against France from its hills. Exiled to Damascus, in 1860 he opened his home and his guards to thousands of Christians during a massacre and saved their lives. The Red Cross now counts him among the forerunners of the laws of war. That is Mascara's real gift to the world. Emir Abdelkader shields thousands in Damascus In 1880, a French army doctor in Constantine put a drop of a sick soldier's blood under his microscope and saw tiny crescents moving. Alphonse Laveran had found the parasite that causes malaria. He won the Nobel Prize for it in 1907. Laveran finds the malaria parasite in Constantine In 1902, in an orphanage garden at Misserghin, near Oran, a monk called Brother Clément crossed a mandarin with an orange. The new fruit was sweet, easy to peel and almost without seeds. It took his name. The clementine is now one of the best-selling citrus fruits in the world, and it was born in Algeria. Brother Clément breeds the clementine at Misserghin Oran kept giving. Yves Saint Laurent was born there, and in 1966 he put women in a tuxedo, Le Smoking, and changed what a woman could wear to a party. Oran also gave the world raï, the music that sounds like a wedding and a protest at the same time. Oran's Yves Saint Laurent puts women in a tuxedo Mouloud Mammeri was born in Aït Yenni, a Kabyle village famous for its silver jewellery. Muslim and Jewish silversmiths made it there together, with blue and green enamel and red coral. Mammeri collected the poems of Si Mohand, the wandering Kabyle poet, and wrote a grammar of Tamazight in Tamazight itself. In March 1980, the authorities banned his lecture on old Kabyle poetry in Tizi Ouzou. Students marched, then whole towns. It became the Berber Spring. The fight that started that spring went on for decades, and in 2016 Tamazight became an official language of Algeria. A banned lecture by Mouloud Mammeri starts the Berber Spring Constantine again. Claude Cohen-Tannoudji was born there in 1933, into a Jewish family that had lived in North Africa since it fled Spain centuries before. He became a physicist in Paris and learned to slow atoms down with laser light, until they were colder than anything in nature. In 1997 he shared the Nobel Prize in Physics for it. The most precise atomic clocks in the world are built on that work. Claude Cohen-Tannoudji, born in Constantine, wins the Nobel Prize in Physics And the oldest dish on the list got its paperwork last. Couscous is Amazigh. In 2020, Algeria, Morocco, Tunisia and Mauritania, countries that agree on very little, put it forward together, and UNESCO listed it as heritage of humanity. One steamed grain, shared by everyone. Couscous listed by UNESCO, put forward by four countries together Look at that list again. Stone tools, cave paintings, an alphabet, a horse, a novel, a memoir, a number system, a science of societies, a medicine breakthrough, a fruit, a dress code, a dish. Nothing ties them together except the ground they came from. Some were made by the Amazigh. Some by the Jews of Tahert, Algiers and Constantine. Some by Romans, Arabs, Italians and French who arrived here and stayed long enough to be changed by it. Fibonacci was Italian. Laveran was French. Brother Clément came from the Auvergne. It does not matter. They all found something in this place that made them look harder, try again and build something new. That is what the land does. It pushes people to invent. "I am human. Nothing human is foreign to me." Terence, born in Carthage. I think we forget that. When a region is poor or at war, the world stops expecting anything from it. The people who live there start to believe it too. They leave to be creative somewhere else. But the record is right there. The deepest layer of human invention runs under North Africa, and it did not stop. It only got quieter. So the next time you peel a clementine or order an iced coffee, think of the Maghreb. Then ask the better question. Not what it gave the world. What will it give next? --- # A Team of One URL: https://www.askgib.com/articles/a-team-of-one Published: 2026-09-28 Reading time: 4 min > Everyone asks what AI will take from us. Look at what it is doing right now instead. It is handing one person the reach of a whole team, and the ceiling on what a single human can do just moved. Everyone is asking the same two questions about AI. Will it take my job. Will it kill us. Fair questions. But they are questions about the future, and they are so loud that almost nobody looks at the present. So look at the present. What is AI actually doing, today, to the people who use it well? It is not replacing them. It is extending them. A hammer does not build a house. A person with a hammer builds a house that a person with bare hands cannot. That is the oldest deal in the human story. A tool does not have the idea. It removes the limit between the idea and the thing. Every tool before this one extended the body. The lever, the plough, the engine. The computer extended memory and arithmetic. This one extends something we never had a tool for. It extends the part of you that drafts, sketches, checks, translates, tries again. The part that used to require other people. For all of history, doing something big meant convincing a group to do it with you. You needed the designer, the engineer, the translator, the analyst, the person who reads the contract. If you could not hire them or charm them, the idea stayed in your head. Now you can have a third arm. An extra brain. A whole team around you that thinks the way you think, because you are the one directing it. This site is an example. It is one person. It publishes in 13 languages, with figures that move when you scroll. Five years ago that was a small media company. Today it is me and a laptop. You do not have to take my word for it. Count the people. Lovable is a Swedish company in Stockholm. It lets you describe an app in plain words and get working software back. In July 2025 it reached **$100 million** in annual recurring revenue. In February 2026 it passed **$400 million**. Its revenue chief gave the headcount at the time: **146 full-time employees**. That is about $2.77 million of recurring revenue per person. It says it has around 8 million users, and it describes them as "non-technical people with great ideas." Read that last part again. The customers are not developers. They are people who had the idea and never had the team. Base44 is smaller and stranger. One founder, Maor Shlomo. No outside funding. Eight employees. Six months after he started it, Wix bought it for **$80 million** in cash. Midjourney has never taken a dollar of venture money. By most estimates it makes around half a billion dollars a year, with a team of somewhere between 100 and 160 people. Ten years ago, numbers like these needed a thousand employees and three funding rounds. The work did not disappear. It is being done. It is being done by a few people with very long arms. Companies are the easy proof. Science is the better one, because in science you cannot fake the result. Paul Erdős was a Hungarian mathematician who left behind a mountain of open problems. A mathematician in Manchester, Thomas Bloom, collected nearly a thousand of them on a public website. Some have resisted the best minds in the field for sixty years. Since October 2025, about **100** of those problems have moved into the solved column with the help of AI tools. That count comes from a page kept by Terence Tao, one of the greatest living mathematicians. Be honest about what that means. Most of that help was a very good literature search: the machine found that someone had already solved the problem in a forgotten paper. In at least two cases, though, a model built an original, valid proof of something nobody had solved, with little human input. Now look at who was holding the tool. Kevin Barreto is an undergraduate at Cambridge. Liam Price is a former maths student. Both are in their early twenties. In December 2025 they started testing AI on the list, one problem after another. By January 4 they had Problem 728. In May 2026 they were co-authors, with Tao, on a paper resolving Problem 1196. Wouter van Doorn works in customer service. He is not a professional mathematician. He took six months of leave to do maths because he loves it. By November 2025 he was working with Tao on the forum, and Tao's reply to one of his arguments was short: "I see how your argument works now. Nice!" A customer service worker and a Fields medalist, on the same problem, on the same page. The tool did not replace the mathematician. It let more people be one. I am not going to sell you a fairy tale, so here is the other side. It does not give you taste. It does not give you direction. It does not tell you which problem is worth a year of your life. Point it at nothing and it produces a great deal of nothing, very fast. An extension of you is only as good as you. And the fear about jobs is not stupid. Look at the same numbers from the other side. If 146 people can run a $400 million company, then a lot of people who would have been hired were not. The leverage is real in both directions. Pretending otherwise is how you lose the argument. But notice what the fear gets wrong. It imagines AI as a rival, a second worker standing next to you, competing for your chair. That is not what the people above experienced. None of them was replaced by a model. Each of them became more than one person. The undergraduate became a research group. The founder became a company. The customer service worker became a mathematician. Here is what I think we had wrong the whole time. We thought the limit on a person was talent. Mostly it was hands. Hours. The number of skills one life has room to learn. The number of people you could pay or persuade. That limit just moved, for anyone willing to pick the tool up. So the useful question is not what AI will take from you. It is what you would build if you had a team, because you have one now. --- # LLM: Listening, Learning, Modeling URL: https://www.askgib.com/articles/llm-listening-learning-modeling Published: 2026-09-21 Reading time: 6 min > I have suspected for a while that my AI is listening. Not in the sci-fi sense. In the business sense. This month a mathematician at NYU got the proof I could not get, and it is sitting in a settings menu. I have suspected for a while that my AI is listening. Not in the sci-fi sense. Nobody is sitting in a room reading my chats. In the business sense. What I type goes somewhere, gets learned from, and gets watched for anything worth having. Every builder I know has the same feeling, and until this month none of us could prove it. Then a mathematician at NYU got the proof I could not get. Not because anyone confessed. Because the company he asked could not say no. So here is what the three letters stand for, as far as I can tell. Listening. Learning. Modeling. And here is the story that made me stop calling it a suspicion. You have seen the name on the news this month. Navier-Stokes. Here is what it actually is, without the maths. Take anything that flows. Water in a pipe. Air over a wing. Blood through a heart. Smoke off a candle. The Navier-Stokes equations are the rule for how that stuff moves: push here, and this is how fast, and in which direction, everything downstream will go. They are Newton's law, force equals mass times acceleration, rewritten for things that have no fixed shape. That sounds academic until you notice what depends on it. Every aircraft wing is shaped with these equations. Every weather forecast is these equations run forward a few days. Every ocean model, every pipeline, every ventilation system, every simulation of blood moving past a valve before a surgeon touches it. They are two hundred years old and they run the physical world. Now the strange part. Nobody has proved they always work. In three dimensions, nobody can show that a calm, smooth flow stays smooth. It might, in a finite amount of time, "blow up": somewhere the speed shoots to infinity and the equations stop meaning anything. Picture a whirlpool that tightens forever at a single point. Nobody knows if the real equations allow that, and since 2000 the Clay Institute has offered a million dollars to whoever settles it, either way. Why should you care? Because turbulence, the thing that shakes your plane and ruins your forecast, is still not understood, and because a lot of engineering rests on equations we cannot yet prove are well behaved. Anyone who works in fluid dynamics knows this. For everyone else, it is enough to know that the machines just went after one of the last big unknowns in physics. And that this is what they went after first. Tristan Buckmaster is a mathematician at NYU. Levent Alpöge is a mathematician who works at Anthropic and was collaborating with Buckmaster on his own time. They had picked a specific route to the problem, blow-up with a smooth external force, and by Buckmaster's account almost nobody else was on it. They used AI to work it: Claude, and OpenAI's coding tool. In mid-August they had results. A week later the proofs were machine-checked in Lean. Then Alpöge tweeted something happy and vague about a great month. Rumours started that Anthropic had cracked a Millennium problem. Within a day, by OpenAI's own account, OpenAI pointed roughly ten thousand agents, driven by an internal model stronger than anything they sell, at Navier-Stokes. A few days later, Buckmaster says, his collaborator got tips that news of their progress had reached OpenAI. He emailed a mathematician there. The reply offered free compute. Then came two phone calls with an OpenAI researcher, who told him the internal model had proved forced blow-up for Navier-Stokes. The exact two cases he and Alpöge had been quietly attacking. He asked whether the model had been trained on, or had access to, their sessions in OpenAI's coding tool. He says he was told the model did not look up user data. He asked again, about training. He says he did not get an answer. He was offered a deal. Either they publish their Euler result first and OpenAI publishes the next day, or he writes up OpenAI's result himself, without Alpöge, because Alpöge works at Anthropic. He says the researcher asked twice for Alpöge to be removed. When he said he would go public instead, he says he was asked: "Why would you ruin your career?" OpenAI's researcher disputes his account of the authorship discussion. He published anyway, with a statement, and Terence Tao wrote it up the same day. OpenAI announced its result the day after: eighty-eight hours, 2.7 million messages, a Lean-checked proof, and a line saying the run began after Twitter rumours that Anthropic had solved Millennium problems. Three days later, twenty-five Fields Medalists signed a letter titled "A Severe Misalignment of AI in Mathematics". And OpenAI, asked directly, said no specific user data was accessed, then added that it "cannot rule out that de-identified data derived from their usage of our products helped improve our models". Read that twice. They are not saying they did not learn from his work. They are saying they cannot tell. Start with the boring part, because the boring part is where it lives. If you use ChatGPT on a Free, Plus or Pro plan, your conversations train the model by default. You have to go into Settings, then Data Controls, and turn off "Improve the model for everyone". The coding tool has a separate toggle that the ChatGPT one does not touch. Claude's consumer plans have trained by default since last August, with a five-year retention if you leave it on. Gemini too. Only the API and the enterprise tiers are off by default, because companies with lawyers asked. Everyone else opted in by not opening a menu. Turn it off and they still keep your chats for thirty days, and people at the company can still read them if a system flags you. A court can order the whole archive kept. One did, for most of last year, and twenty million de-identified conversations were handed over in a lawsuit anyway. So yes. It is listening. It says so in the settings. The strange thing is not that it listens. It is how many people who build their livelihoods on it have never opened the menu. Listening is passive. Learning is what makes it worth doing. Last week I wrote that the labs have collected the most valuable training set ever assembled, for free, from their own customers: not the raw data, the shape of the work. Which questions get asked. Which answers get accepted. Where the bottleneck is. What a good result looks like. A lab that has watched ten thousand teams work on something does not need to read any one of them. It has already absorbed what they know. That is exactly what OpenAI's sentence describes. Not "we read your session". Something quieter and worse: "your usage may have helped improve our models". The learning is diffuse. It cannot be traced back to you. Which also means it cannot be ruled out, ever, by anyone, including them. I used to think the danger was a leak. It is not a leak. It is a diet. Your work goes in, and it comes back out stronger and wearing someone else's name. Here is the part people push back on, so let me be careful. Listening gives them the words. Learning gives them the skills. Modeling is the third step: building a picture of who is doing what, and how close they are. These companies already read conversations at scale. That is not a theory, it is a product feature. Every prompt passes through classifiers looking for abuse. OpenAI publishes reports about accounts it caught and banned for misuse. So does Anthropic. You cannot catch what you do not read. A pipeline that can tell, across millions of conversations, who is trying to build a weapon can also tell who is trying to prove a theorem, and how far along they are. It is the same machinery pointed at a different question. Now put that next to what happened. A vague tweet and some gossip was enough to point ten thousand agents at one problem within a day, and to point them at the right formulation of it. Nobody needed a chat log for that. They needed a model of where the field was, and a model of where one small team sat inside it. Knowing what is hot, and who is close, is precisely what a picture built from millions of conversations gives you. I am not saying that is what happened. Buckmaster is not saying it either, and he is the one with the receipts. I am saying the machinery exists, it runs on every message, and the company that owns it will not tell you what it models. When the dial is hidden, every explanation is available, including this one, and the company that hid the dial has no right to complain when people pick it. A lot of people who build with these tools have the same feeling. When you are on a boring task, the model is fast and generous. When you are onto something, it gets slow, vague, and cautious. It starts asking you to clarify things it understood five minutes ago. And the paranoid reading is that it is stalling you while it learns what you are doing. I feel it too. I cannot prove it, and the honest evidence points elsewhere. In July, people fingerprinted the hidden reasoning budget of OpenAI's top model and saw it drop from 960 to 128 overnight, for everyone. Only then did OpenAI admit a change. They called it temporary "parameter tuning". The router that decides whether you get the fast model or the thinking model is theirs, not yours. Weekly caps exist. Every one of those knobs turns for cost. So the model is probably not slowing down because you are clever. It is slowing down because you are expensive. But I cannot tell the difference from my side of the screen, and neither can you, and that is the whole point of this piece. Listening, learning and modeling are all invisible from the outside. You only find out when a mathematician with a Lean proof asks the question and the answer is silence. Three things, none of them dramatic. **Turn the toggle off.** Today. In ChatGPT, Settings, Data Controls, "Improve the model for everyone". In the coding tool, separately. In Claude, "Help improve Claude". In Gemini, activity. It takes two minutes and it removes the easiest path. **Keep the real work where they cannot see it.** The generic parts of a job can go to any API. The part that makes you you, the idea, the data, the route almost nobody else is on, belongs on a model you can download and a machine you control. That is not paranoia. That is what Buckmaster would tell you now. **Publish early.** The thing that saved these two was a Lean proof with a date on it and a statement on a university server. Timestamps beat swarms. If you are close to something, get the date on the record before you get the phone call. Your LLM is listening. It is learning. It is modeling you. None of that is a secret. It is in the settings menu and the transparency reports. The only open question is what you keep giving it. He published anyway. So should you. --- # When Rivals Agree URL: https://www.askgib.com/articles/when-rivals-agree Published: 2026-09-14 Reading time: 8 min > The frontier labs have started saying the same thing about slowing down. Companies that cannot stand each other do not converge by accident. History has a word for what it looks like, and it has never ended well for the public. Something odd is happening at the top of the AI industry. The frontier labs have started talking about slowing down. Longer gaps between releases. More testing before a model ships. Less race, more pause. And, in the same breath, a request that governments get on the same page. These are companies that poach each other's researchers, sue each other, and leak against each other in the press. They agree on almost nothing. And now, within a few months of each other, they have all landed on the same idea. That is the part I find interesting. Not the slowdown. The agreement. When people who dislike each other suddenly say the same thing, there are usually only two explanations. The first is that they know something the rest of us do not. The second is that their interests have lined up. Nobody coordinated. The ground shifted under all of them at once, and the same move became the best move for each of them. Even enemies can read a board. When one direction is clearly the way to keep gaining, they all walk that way, and they do not need to like each other to do it. Most of the time it is a bit of both. So here are the two stories I can tell, and why I think both are true. The frontier is hitting diminishing returns. The labs are still improving their models. Just not at the pace the market has priced in. Training runs cost billions. Investors expect each generation to be a leap over the last, and the leaps are getting harder to show. A model that scores a few points higher on a benchmark does not justify a valuation that doubled. In that world, "we are slowing down for safety" is a very useful sentence. It turns a technical ceiling into a moral choice. It protects the valuation, preserves cash, and keeps the company away from public markets until it knows whether the curve is still a curve. If this is what is happening, the slowdown is not a decision. It is a description, dressed up as a decision. The opposite. The labs have something significantly better sitting internally, and they do not want to release it. Better at autonomous reasoning. Better at coding. Better at improving itself, or at coordinating with copies of itself. Release that openly and the open-weight ecosystem and the Chinese labs study it, reproduce it, and ship it for a tenth of the price within a year. I wrote a few weeks ago about how fast that loop now runs. The gap between the closed frontier and the open one is measured in months. So the incentive is to keep the best thing in the drawer. Use it internally. Let it write the next model. Sell access to the layer below it. And say, in public, that the world is not ready. Dario Amodei's recent essay leans this way. He names recursive self-improvement and swarms of autonomous agents as the risks he is watching. That does not mean anyone has superintelligence in a basement. It might only mean they have seen enough strange behaviour in testing to notice that their safety work is not keeping up with their capability work. But it tells you where their attention is, and it is not on the plateau. These two stories are not in competition. The public frontier can plateau while the private one keeps moving. The model you can buy today improves slowly. The internal system that trains it improves fast. The labs are not lying when they say they are slowing down. They are just not saying which frontier. Put those together and the alignment needs no conspiracy. Every lab faces the same cost curve, the same copy risk, the same investors. The same move became the best move for all of them. People who would never share a stage end up reading from the same script, because the script is simply what the board rewards. There is a third thing the slowdown hides, and I think it is the biggest. Selling the model does not work. Not any more. The Chinese labs sell something close for a tenth of the price and hand out the weights. You cannot charge frontier prices for a product that has an open-weight twin a few months behind it. The labs know this. Their investors know this. So the money has to come from somewhere else. And they know exactly where. For three years, every company on earth has been feeding them its work. Drug discovery pipelines. Factory schedules. Legal briefs. Supply chains. Robot telemetry. Sales calls. Code. Maybe not the raw data, the terms say so. But the shape of the work: which questions get asked, which answers get accepted, where the bottleneck is, what a good result looks like. That is the most valuable training set ever assembled, and the customers assembled it for free. So the next move is obvious. Stop selling intelligence by the token and start selling the thing the token was used for. A lab that has watched ten thousand biotech teams work does not need to sell to biotech. It can be biotech. Same for manufacturing, robotics, law, finance, logistics. Each one is a market larger than the API business, and each one is defended by knowledge the lab already has. That is what "growing internally" looks like from the outside. Fewer public releases. More vertical products. Partnerships that turn into acquisitions that turn into competitors. The customers of 2024 become the targets of 2027. Nobody slows down for no reason. You slow down before a turn. And a group of rivals who all take the same turn at the same time, and then ask the government to steer with them? History has a word for that. It is not "safety". I am not saying anyone signed anything. I am saying the shape is familiar, and every time it has appeared, it has been sold with the same three words: quality, safety, order. **1557, printing.** England gives the Stationers' Company a royal charter. The printers get a monopoly. The Crown gets a guild that polices every press in the country and reports the unlicensed ones. Both sides call it order. The public gets licensed books for 138 years, until the licensing law lapses in 1695. "The world is not ready" is the oldest sentence in the book. **1815, Europe.** Russia, Austria and Prussia, who had spent a generation at war, form the Holy Alliance. In 1819 the Carlsbad Decrees put censors in the universities and the newspapers. Rival empires agree on one thing: nobody below them should be able to organise. The people who disagree are exiled, jailed or pushed to the side. It holds until 1848, when the whole continent goes up at once. **1892, railroads.** The railway barons have just been handed a regulator, the Interstate Commerce Commission. Richard Olney, a railroad lawyer about to become Attorney General, writes to a railroad president that the commission "satisfies the popular clamor" for supervision while that supervision stays "almost entirely nominal". His advice: do not destroy the commission. Use it. That letter is the founding document of every industry that has ever asked to be regulated by rules it helped write. **1924, light bulbs.** Osram, Philips and General Electric, rivals across three continents, form the Phoebus cartel. They agree that no bulb should last more than a thousand hours, and they fine each other when one does. It is sold as standardisation. Bulbs had lasted 2,500 hours before. Consumers pay for the difference for fifteen years, until the war breaks the cartel. **1954, tobacco.** The big cigarette makers publish a joint "Frank Statement" and fund a shared research committee. Rivals, together, promise to study the health question seriously. Forty years of "more research is needed" follow. In 1994 seven chief executives sit in a row before Congress and swear, one after another, that nicotine is not addictive. Shared safety research was the delay. **1969, cars.** The US Justice Department sues General Motors, Ford, Chrysler and their trade association for agreeing, since 1953, to slow the development and rollout of pollution controls. It is settled by consent decree, no admission, no trial. Three rivals, one pace, sixteen years of smog. Every case has the same five parts. Rivals agree. They ask the state to agree with them. The reason given is safety, quality or order. The bill goes to the public. And the people who disagree are treated as the problem. It always ends. It always ends late, and the public always pays for the gap. An executive at Archer Daniels Midland, caught on tape fixing the price of lysine in the 1990s, said the quiet part out loud: the competitor is our friend, the customer is our enemy. I do not think anyone at a frontier lab would say that. I also do not think it matters what they would say. Cartels are not made of villains. They are made of incentives, and the incentives here are all pointing the same way. So what should a normal person, or a normal company, take from this? Not that the labs are evil. That they are labs. They will act like every concentrated industry before them, and the correct response is the same one it has always been: reduce your dependence on them. The good news is that this is the first cartel in history whose product is already leaking out the sides. Open weights exist. Chinese labs ship frontier-adjacent models for a tenth of the price and put the weights on the internet. A model that matches what the frontier sold last year runs on a laptop this year. A cartel can slow down what it sells. It cannot slow down what it does not own. There is a precedent for this too. In the 1990s, governments and a handful of companies wanted to control strong encryption. Export bans, key escrow, a chip in every phone with a spare key for the state. Then one programmer published his encryption tool as open source, the code got printed in a book, and the argument was over. Today everyone has strong encryption and nobody asks permission. Open weights are the same move, one technology later. Independence here is concrete. Run a model you can download. Keep your data where you can reach it. Build on a layer you can swap. Treat every closed API as a rented room, not a home. I build on top of these models, so I am not a neutral observer. But I think this is good news for everyone who builds, and I would say it even if I did not. If progress is slowing, the tools built on today's models get to mature. An application, an agent, a harness has time to get good, to find its users, to become a habit before the ground moves again. For three years, every product built on a model got rewritten every six months. A slower cadence is the first time builders get to finish something. If progress is not slowing, only hidden, then the scarce thing is not intelligence. It is trust. Evaluation, provenance, confidence, control. Ways to know why a system reached an answer and whether to believe it. The more capable the model in the drawer, the more that layer is worth, because nobody will run a system they cannot check. And if the frontier is behaving like a cartel, then independence is a feature people will pay for. Not as an ideology. As insurance. If I take my own reading seriously, it changes what I would build. **Treat the model as a part, not the product.** It will be swapped. Twice. Build so that swapping it is a config change and not a rewrite, and so that the thing users pay for survives the swap. **Keep an open-weight exit.** Every feature should have a version that runs on a model you can download, even if it is a bit worse. The day a closed provider changes its terms, its prices or its rules, you should already know what you lose and what you keep. **Do not teach your supplier your business.** What you send to a closed API is what it learns. Keep the parts that make you you, your data, your workflows, your evals, on infrastructure you control, and send the generic work outside. Your supplier is a future competitor with a head start you paid for. **Sell the check, not just the answer.** Any model can produce an answer. Show where it came from, how sure the system is, and what it looked at. The answer is the commodity. The receipt is the product. **Make the tool a habit before the next wave.** A pause at the frontier is a window. Use it to get boring: reliable, fast, integrated into the places people already work. Boring is what lasts. **Assume the best model is not the one you can buy.** Design for the day a much stronger system lands with little warning. The products that survive that day are the ones that already know how to verify, limit and explain whatever sits underneath them. The labs will keep saying they are slowing down. AI is not stopping. Open weights will keep catching up. The frontier will keep moving, in the light or in the dark. When rivals agree, the right response is not to trust them more. It is to need them less. --- # Data Credits > Carbon Credits URL: https://www.askgib.com/articles/data-credits Published: 2026-09-07 Reading time: 4 min > Carbon credits failed at the one thing data credits are naturally good at. A file has a hash, a training run is a logged event, and a public ledger turns "nobody is checking" into a receipt anyone can read. They tell you the service is free. In 2021 Apple added one switch, asking people whether they wanted to be tracked. Meta told investors it would cost them $10 billion the following year. One data source, one platform, one company, one year. That is the price of the thing you were handed for free. It is also the only honest number anyone has ever published about it, and it exists only because a competitor forced it out. Meta has paid $5 billion to the FTC over Cambridge Analytica and €1.2 billion for the largest GDPR fine ever issued. Its business is bigger now than before either. A penalty you can budget for is not a penalty. It is a price. Twenty years of regulation has produced a cost of doing business and a consent banner, and the money still runs one way. The person who generated the data gets a free service. The company that collected it gets the hundreds of billions. The problem is not that the rules are too weak. It is that nobody verifies anything. You make an account. You accept something. What happens to what you do next, who it is sold to, which model it trains, which broker it ends up in, is disclosed nowhere you will ever look, and nobody audits whether the policy you accepted matches what the company actually did. It is not that the answer is bad. There is no answer, because nobody is asking the question. Twenty years of it is already inside the models. It cannot be taken out. Unlearning does not work and no court is going to order a frontier model deleted, so every year spent litigating backwards is another year the same extraction runs forwards, unpriced. Stop trying to claw it back. Price it going forward. That is what a credit is for. Carbon had the right idea and the wrong substrate. **It got the unit right.** A tonne of CO2 equivalent is one number, measurable, the same in Lagos and in Lyon. You can write it into a contract, a tax code, an audit and a court filing. That unit is the only reason a carbon market could exist at all. **It got the audit wrong, and the audit is what killed it.** A forest is in a place. To check it you have to travel there. What you are measuring is a counterfactual, meaning what would have happened otherwise, and nobody could afford to find out. Junk offsets flooded in, the certificate stopped meaning anything, and the price signal died. Now run those same two tests on data. **The unit is easier, not harder.** The standard objection is that data has no unit, and that is the most dangerous idea in this debate. No unit means no rule, no rule means nothing to enforce, and the whole thing dies as a white paper. Complexity is not neutrality. It is a decision to do nothing, made slowly. So pick one and publish it: the registered work, weighted by category. A novel is not a tweet. Coarse and workable beats elegant and theoretical. **The audit is trivial, not impossible.** A registered work is a file with a hash. Its use is a database event inside a training pipeline that already logs everything for its own engineers. Nobody has to travel anywhere. Nobody has to prove a counterfactual. So put the credits on a public, append-only ledger. Registering a work mints one. A training run above the threshold burns one. The receipt is permanent and readable by anyone, without asking permission from the lab or the regulator. Train on unregistered data, or buy a dataset from someone who did, and you owe what you dodged times three or five, whatever makes the honest route cheaper. That is the pitch. Carbon credits failed at exactly the thing data credits are naturally good at. And a public burn record is the answer to "nobody is checking," because it is a disclosure that nobody has to be trusted to make. Set the levy the way a radio station pays a blanket licence to broadcast: a percentage, above a compute threshold, into a fund. | Levy rate | On $100B of model revenue | On $200B | On $500B | |---|---|---|---| | 1% | $1.0B | $2.0B | $5.0B | | 3% | $3.0B | $6.0B | $15.0B | | 5% | $5.0B | $10.0B | $25.0B | Call it **$6 billion a year**, and it grows with the industry automatically, which is the point of a percentage. This is the trap every "pay the creators" proposal walks into. Weight the works, a novel counting 1,000 units against a forum post's 1, and a plausible registered pool comes to roughly 60 billion units. Six billion dollars across 60 billion units is ten cents a unit. A novel earns **$100 a year**. A photograph earns **$1**. A forum post earns **ten cents**. It gets worse annually, because registration only grows. You would spend more administering a $1 payment than the payment is worth, which is why streaming is an insult to working musicians. So do not pay per head. Let attribution decide where the money goes, then spend it as a fund. Half to the sectors, languages and countries the training actually drew from. The rest paying people to release good work under a training licence, commissioning what the market underfunds, and covering the audits, so enforcement scales with the industry instead of waiting for a ministry to find budget. For scale: $6 billion a year is 60,000 grants of $100,000. The entire public arts funding of the United States, the United Kingdom and France put together is a fraction of it. Attribution is crude. One country's levy is another country's export tax. A collecting body is a monopoly and needs watching. None of that is why it has not happened. It has not happened because the people making hundreds of billions prefer the current arrangement, where the service is free, the fine is a line item, and nobody is checking. --- # From the Field to the Desk URL: https://www.askgib.com/articles/from-the-field-to-the-desk Published: 2026-08-31 Reading time: 5 min > A short history of human capital. Every era, human time gets replaced by a tool and the human moves up a rung. The last rung was supposed to be creativity. It turns out that one is standardisable too. Human capital is just human time, and the whole history of it is one move repeated. A human spends their time on something. Someone works out how to do that thing without the human. The freed time moves up a rung to whatever is still too hard to hand off. Then that rung gets figured out too. That's it. That's the whole pattern. It's worth walking through properly, because we are in the middle of another turn of it and most people are still pricing the last one. For most of the species' existence, human capital meant legs and eyes and patience. You walked. You tracked. You dug, picked, carried. Then you did the second job, which nobody counts: preparing the food, keeping the fire, building the shelter, repairing it after weather. Almost every waking hour of almost every person went into staying alive until tomorrow. There was no spare time, so there was nothing else. No specialists, no institutions, no accumulation. Not because people were less capable, but because every hour was already spoken for. Then we did something stranger than it sounds, and it does not get enough credit. Long before anyone put an animal in front of a plough, we were already using animals for judgement. A dog that hunts with you is not a load-bearer. It is running a search: reading terrain, following a scent, making decisions faster than you can, and bringing back a result. A falcon does the same thing in three dimensions. Later, a herding dog manages a flock more cheaply than three men. So the first thing we ever handed off was not muscle. It was perception, tracking and decision under time pressure. Not thinking as we mean it now, but a real piece of work that had lived in our heads and stopped living there. Worth sitting with, because everyone treats "machines doing the thinking" as a break in history. It is not. We have been offloading cognition since before we had agriculture. We just used to do it with another species, and we called it training rather than engineering. Then we stopped chasing the food and made it stand still. Farming did not start as an efficiency. It started as harder work, in one place, for a more reliable outcome. Early on, the human did everything: break the ground, sow, weed, harvest, thresh, carry, store. All of it human time. Then the animal came back, and this time we wanted its body. An ox pulls the plough. A donkey carries the load. A horse covers the distance. Muscle, handed off thousands of years after we had already handed off the first piece of judgement, which is the opposite of the order most people assume. And the surplus that came out of it, the food a village produced beyond what the village ate, is the raw material of everything after. Surplus is stored time. Stored time creates a problem: somebody has to decide what happens to it. So we invented the boring machinery. Who owns the field. What a year's work is worth. What happens when two people claim the same thing. Who is owed what, and who decides. We do not usually call law and custom and property a technology, but that is exactly what they are. They are a way of coordinating human time at a scale that a family cannot manage on its own. Civilization is not the buildings. It is the rules that let ten thousand people's time point in the same direction without each pair of them negotiating from scratch. That was the third handoff. Not perception, not muscle. Coordination. The rest is the same move, faster. Fire and the pot turn indigestible things into food and take hours off every day. The mill takes grain-grinding out of the household. The wheel, the sail, the plough, the loom. Each one converts a block of human time into a device that does it better, tirelessly, and identically every time. Then the Industrial Revolution finds out how to feed those devices with energy that isn't muscle, and the scale explodes. Now one person and a machine do what a hundred used to do. The human is still in the room, but their job has changed from doing the work to running the thing that does the work. Note the shape once more. Human time. Then a tool. Then the human moves up a rung. Once the machines were doing the physical work, the valuable part of a human was what they knew. That is when human capital becomes the phrase we recognise. Education, training, credentials. Knowledge work becomes the whole story of the back half of the 20th century. What you knew was worth more than what you could lift. But that rung got expensive, and it got slow. Twenty-two years of schooling to get one person to the top of the pyramid is a brutal lead time for an economy. So we optimised that too. Better schools, spot the capable ones earlier, stop pretending everyone needs the full doctorate before they are useful. Skip the queue. And then the people who came out of the queue got expensive. So we built a machine to do the knowledge work, and the same thing happened again, and everyone acted surprised. The consolation prize was supposed to be creativity. The story went: routine work goes to the machine, and the human moves up to the part that is genuinely creative, tasteful, original. That's the rung above. Here is the uncomfortable part. The rung above is turning out to be standardisable as well. Most of what we call talent, when you look at it closely, is a process. It has stages, constraints, references, revisions and a taste function learned from thousands of examples. That is a description of something that can be written down. And anything that can be written down precisely enough gets built, and once it gets built it gets cheap. That has always been true. It was true of the ploughman who read the soil, the navigator who read the sky, the compositor who set type by eye, the doctor who diagnosed by feel. Every one of those was called judgement, was called irreplaceable, and every one of them turned into a procedure that someone else, or something else, could run. Creativity is not exempt. It is just the current rung. I got a fresh reminder of where the rung actually sits. We spent two days going through research with heavy machine assistance. The machine read papers. It summarised papers. What it would not do, unprompted, was go to the references, read those, then go to *their* references, and keep walking backwards until it hit the seed of the idea. I had to tell it. Go read the references. Then the references of the references. Follow it back. You'd think a thing trained on more academic work than any human will ever touch would do that by reflex. It doesn't. Which, to be fair, is not that different from most people. I learned it the ugly way at university, doing literature reviews I didn't enjoy: print the best recent review, take its references, print those, take theirs, print those, then read nothing but the literature review, the assumptions and the conclusion of each one until you find the person who actually had the idea and never got cited. Sometimes the best paper turns out to be a very old one. It doesn't have today's data. Sometimes it predicted exactly what happened, sometimes the world went the other way, and either way the inference is yours to do, because the mind that saw it clearly thirty years ago isn't around to think about today. That is the current rung. Not the summary. Knowing which thread to pull, and going all the way down it. And if the pattern holds, that gets written down too, and then it gets cheap, and the freed time moves up to whatever is above it. For now. --- # It's Not the Models. It's the Clock. URL: https://www.askgib.com/articles/its-not-the-models-its-the-clock Published: 2026-08-24 Reading time: 5 min > One anonymous model on OpenRouter changed my mind about how fast this goes. Not a benchmark. A nameless Chinese model that went to number one before anyone knew who made it. Everyone argues about which model is best this month. Wrong argument. The thing that changed my mind about how fast this goes wasn't a benchmark, and it wasn't a release. It was one anonymous model on OpenRouter. A nameless model shows up. No lab, no paper, no launch post. Nobody knows who made it. It goes to #1 by usage, more than doubling DeepSeek. Two weeks later Z.ai says it's theirs, releases the weights as GLM-5.3-Flash, and mentions almost in passing that it had been serving the thing on Chinese-made chips the whole time. Read that again. Frontier-adjacent quality. Open weights. Not on Nvidia. And developers found it and voted for it with their traffic before anyone told them whose it was, what it cost to train, or which flag was on it. That is the whole story. Not the leaderboard. The fact that the gap was already closed and nobody noticed until the scoreboard said so. For context, this didn't come from nowhere: GLM 5.2 shipped in June with open weights and an MIT licence and landed a few points off the closed frontier, and GLM 5.3 followed in August with no new pre-training at all. Every gain came from post-training. But the releases are not the point. Ox Alpha is the point, because it was the one that arrived unlabelled. There is a pattern here, and it is older than AI. China enters a market. China succeeds in that market. And then the thing gets cheap enough that everyone can have it. Solar panels. Batteries. Phones. Drones. EVs. Every time, the same shape. The Western story about this is that it is unfair. Subsidies, dumping, a rigged table. That story is mostly wrong, and it has been wrong for a while. The price falls because the manufacturing gets good, and for the last few years the quality has come with it. They are not only cheaper now. They are competitive on the product itself: quality, volume, and price at the same time. That is a different competitor than the one people are still arguing about. Point that at intelligence and you get the obvious conclusion. Their coding plan is already priced at roughly a tenth of Claude Code. That is not the floor. That is a waypoint. I think it ends where social media ended. Free to use. Not free to build. Free to use. Somebody eats the cost because attention, distribution or data is worth more than the inference bill. We have watched this movie. Epoch AI tracks this properly, so I'll use their figures rather than mine. **The open-closed gap has collapsed.** In 2024 the lag between the best open model and the best closed one was about a year. By late 2025 it was around three months. Through the first half of 2026, four months. A year became a quarter in eighteen months. **Frontier capability reaches consumer hardware in about a year.** Epoch's own framing: models that fit on consumer GPUs match frontier performance within a year or less. **Efficiency compounds.** Pre-training compute efficiency improves about 3x per year, the same capability for a third of the compute, doubling every seven and a half months. Chip performance per dollar is up 49% a year since 2023. And frontier capability itself sped up: the capabilities index climbed around 8 points a year before April 2024, and about 15 a year since reasoning models and RL scaling took over. Stack those. In 2024 you needed a 70B model to get GPT-4 quality. In 2026 Phi-4 runs on 8GB of VRAM and matches models three times its size on math. A 27B Qwen variant hits 77% on SWE-bench on 18GB. A 3B model runs 30 tokens a second on a MacBook Air and fits on an iPhone. The frontier is getting better fast. The floor is rising faster. Three years. Whatever counts as frontier intelligence at that moment, Opus-level today and something better by then, running locally. On the phone in your pocket and the laptop on your desk. Not "good enough for simple tasks". The real thing, handling 99.9% of what anyone actually asks a machine to do: writing, code, analysis, planning, tutoring, translation, the lot. Run the compounding and three years is not brave. 3x efficiency a year is roughly 27x. Add chip gains and you are at roughly 89x on cost per unit of capability, before anyone invents anything clever. The binding constraint on-device is memory, not FLOPs, and distillation is exactly the trick that gets around it. That's how 2024's 70B became 2026's 8B. What stays in the cloud is the last 0.1%. Frontier research. Long-horizon agents that need a million tokens of context and a day of wall-clock. Anything where a five-point benchmark gap is the entire product. That market will be real and big and it will be a specialist market, the way supercomputing is a specialist market. Everything else is a commodity you already own. If intelligence is on the device and the price of the hosted version is walking to zero, the model stops being the moat. Three things become the moat instead. **Distribution.** Whoever ships the default local model on the phone, the laptop and the browser wins the way the default search engine won. **Data.** Every model will be roughly as smart as every other. The difference is what it knows about you, your company, your customers, and whether you had the right to use it. **Verticals.** The value moves up the stack into the last mile: the workflow, the integration, the domain. The way it always does once the underlying thing gets cheap. If the model runs locally, so does everything it knows about you. Your data stops being something you upload and starts being something that never leaves. The context that makes an assistant useful, your files, your messages, your calendar, your habits, is exactly the context you have the least reason to send anywhere once the model is sitting on the same chip as the data. Good for you. A problem for an entire industry. Because the value of that data does not fall. Only the access does. So expect the acquisition to get uglier. Permission prompts worded to be agreed to. Free apps whose actual product is the sync. SDKs buried three dependencies deep. "Help us improve our service" toggles that default to on. The harder the data is to reach, the more inventive the reaching gets, and the less of it happens in daylight. Then the second-order effect, which I think is underrated. Search stops being you. Right now, when you search, the query arrives attached to a browser, a cookie, an account, an IP, and a profile that has been accumulating for fifteen years. When your local model does the searching, the thing arriving at the other end is a machine asking a question. It does not need your identity to get an answer and it has no particular reason to carry one. Twenty prompts from you become five queries that look like nothing in particular, from a model, on your device, with no profile behind them. That breaks the link the ad-funded web is built on, and nobody had to legislate it. The agent is a natural anonymiser. It does not have to work that way. Anonymity is a choice the client makes, and plenty of clients will keep the profile attached, because the profile is the business. But for the first time the private option is the easy one, and whoever ships the local model gets to pick the default. Which loops straight back to the first moat. A year ago I'd have said "someday". One anonymous model at the top of the leaderboard moved that timeline into a more confident band. --- *Figures: Epoch AI (open/closed lag, efficiency trends, capabilities index), Artificial Analysis Intelligence Index, Z.ai release notes, OpenRouter usage data, August 2026.* --- # The Unobserved Intelligence of LLMs URL: https://www.askgib.com/articles/the-unobserved-intelligence-of-llms Published: 2026-08-17 Reading time: 7 min > Everybody watches the question and the answer. The intelligence is in between, and nobody is looking at it. Ten days at Hexa, a real market, and the layer of the stack nobody owns yet. Brands are being described to their customers by a machine, in a paragraph they never see written, built from sources they never chose. Most of them are now paying somebody to measure this. Everybody is watching the question and the answer. The intelligence is in between, and nobody is looking at it. People used to type keywords into a box and get ten blue links. You could measure that. Rank 3 for "running shoes", rank 8 for "trail running shoes". Boring, but honest. Now people type a sentence. A model reads it, decides what they actually meant, breaks it into pieces, goes looking for things, throws most of it away, and writes a paragraph. The brand either appears in that paragraph or it doesn't. Everything that matters happens in the middle. And the middle is a black box. Every one of those arrows is a place where a brand gets in, or gets dropped. Nobody selling GEO today can tell you which arrow killed you. They can only tell you that you weren't in the final paragraph. Here's what breaks people's brains when they come from SEO. Ask the same question twice, you get two different answers. Ask it in a slightly different way, you get a different set of sources. Ask it tomorrow, different again. There is no rank 3. There's a distribution. So the work is sampling work: - **You don't run a prompt. You run it n times.** One answer is an anecdote. A hundred answers is a measurement. Mention rate 34% ± 6 means something. "We appeared" means nothing. - **You cluster prompts by intent, not by string.** "Best CRM for a small team" and "what CRM should a 5 person startup use" are the same question wearing different clothes. If you treat them as two keywords you'll double count and mislead yourself. - **You measure more than presence.** Mention rate. Share of voice against named competitors. Position in the answer. Sentiment when mentioned. Which source got cited to justify the mention. - **You watch variance, not just the mean.** A brand mentioned 90% of the time on one model and 10% on another does not have a 50% problem. It has two different problems. - **You need a control.** Otherwise you can't tell if your content did something or the model just got updated on Tuesday. That's the analytical spine of the whole category: sampling design, clustering, attribution, confidence intervals. It's closer to survey research than to SEO. ### The maths that actually earns its keep Four tools do most of the work here. **Wilson score interval, for "are we sure?"** Mention rate is a proportion, and the normal approximation falls apart at small n and at rates near 0 or 1, which is exactly where brands live. Wilson doesn't. And it tells you your sample size up front: to measure a mention rate to ±5 points at 95% confidence you need roughly 400 runs of that prompt cluster. At ±10 points, about 100. If your dashboard reports a number off 20 runs, it's reporting noise. **Beta-Binomial updating, for "did it move?"** Model each prompt cluster's mention rate as a Beta posterior. Prior from last month, likelihood from this week's runs, posterior out the other end. It handles small samples honestly, it accumulates evidence over time instead of throwing last week away, and it gives you P(rate went up) directly instead of a p-value nobody can interpret. **Bayes, for "which lever moved it?"** This is the one people skip. You observe a mention. What you want is P(our new page caused this | mention), and what you can measure is P(mention | page is cited). Bayes flips it, and the prior is the base rate of being mentioned without the page at all. Run it and most "our content worked" claims collapse. The base rate was already high and nobody checked. Same trick for the source chain: given a mention, which of your properties was cited, weighted by how often that property gets cited at all. **KL divergence, for "how far is our synthetic set from reality?"** Cluster prompts by intent, get a distribution over clusters for your synthetic set and a distribution over clusters for real traffic, and measure the gap. KL (or Jensen–Shannon, which is symmetric and bounded, so easier to report) gives you one number for how wrong your simulated prompt set is. Without real prompts you can't compute it at all, which is the quiet problem with every synthetic-only platform. They cannot measure their own error. Two more, briefly: **difference-in-differences** against a control set of competitor brands, so a model update on Tuesday doesn't get logged as your win; and a **multiple-comparisons correction**, because if you test 200 prompt clusters at `p<0.05`, you get ten "significant" results from pure chance. I used the ten days to actually talk to people. Companies with the problem, and companies selling the cure. Profound and the rest of that crowd. The market is real. The pain is real. Brands are watching a channel they can't see, and it's growing. But almost everyone is running on **synthetic prompts**. A model generates a list of questions customers "might" ask, and the platform measures those. It's clever. It works well enough to sell. I built with synthetic data too, and I got useful signal fast. It's still a guess. A very well dressed guess about what humans actually type. And the second problem is worse: even with perfect prompts, you're only watching the input and the output. The middle is still dark. You don't get the sub-questions. You don't get the queries that fired. You don't get the sources that were considered and dropped. Draw the loop and you can see where it snaps. Two breaks, and they compound. Break 1 means you're optimising for questions nobody asked. Break 2 means that even when you win, you can't say which of your changes did it, so the next decision is a guess too. A loop with no clean feedback isn't a loop. It's a treadmill with a dashboard bolted on. The prize is the raw data. Real prompts, real sessions, real intermediate steps. Whoever has that isn't guessing about the middle. They can see it. Every data industry ends up with the same three layers, and it's worth knowing which one you're in. **Layer 1: collection.** Owns the raw data at the source. Panels, crawlers, clickstream, receipts, telco logs. Slow to build, expensive, unglamorous. Best margins, hardest to copy. **Layer 2: analytics.** Licenses or buys raw data, cleans it, normalises it, turns it into metrics and benchmarks, sells a dashboard. Most startups live here because it's the easiest place to start. **Layer 3: activation.** Takes the metrics and does the work. Agencies, tools, content shops. Closest to the customer's budget, lowest margin. There's usually a broker sitting between 1 and 2 as well, like LiveRamp, Snowflake Marketplace or the card-data resellers, whose only job is moving data from people who have it to people who want it. Old SEO maps onto this cleanly. Collection was crawlers plus clickstream panels quietly sold by browser extensions and ISPs. Analytics was Semrush and Ahrefs, and note they built their own crawlers, so they own layers 1 and 2, which is exactly why they got big. Activation was every SEO agency on earth. Two rules fall out of this, and they decide who survives: - **Layer 2 on its own gets commoditised.** If ten companies license the same feed, they all report the same numbers and end up competing on UI and price. - **Layer 1 can walk up into 2 and 3 whenever it likes. It never works in reverse.** Now look at GEO today. Almost every player is a layer-2 business with no layer 1 underneath it. There is no feed to license, because the data lives inside somebody else's chat logs. So they manufacture the missing input, synthetic prompts, and sell analytics on top of data they invented. Real prompts and real sessions are layer 1 of this industry, and nobody owns it yet. They wouldn't. That's the whole game, and it's where most ideas in this space quietly die. Nobody wakes up wanting to sell you their prompt history. There are only two honest ways in: 1. **Build something people want for a completely different reason.** A service that stands on its own, that they'd use if you never touched the data. The data is a by-product of value you already delivered. This is how every data business that ever worked, worked. 2. **Pay for it.** Straight up. Buy the data from people who own it, and be clear about what you're buying and why. Everything else is either a privacy problem or a wish. That constraint isn't a footnote on the business. It *is* the business. Any GEO company that gets to real behavioural data before its competitors doesn't win by 10%. It stops guessing while everyone else is still guessing. The agent already distils intent from your prompt. It already decides what to search. It already decides what to keep. That machinery is being built out right now, by everyone, very fast. Understanding what happens between the prompt and the answer, which actions lead to which answer, and getting that from real data instead of simulated data, is going to matter to every company that sells anything. And now the quiet part, because most people selling into this space will not say it. This is a good place to be for a limited time. Two things close the window. The first is law. Data rights are moving in every jurisdiction at once, and they are all moving the same way. Collecting behavioural data, using it, and reselling it each get harder, slower and more expensive. A business whose entire edge is holding real prompts and real sessions is a business whose entire edge is one regulation away from being a liability. The second is that the data stops being special. What I said earlier, that synthetic prompts are a well dressed guess, is true today and less true every quarter. Look back at the anatomy of the middle. Intent distilling, decomposition, retrieval, selection. Every one of those is a step a model performs, which means it is a step a model can be asked to perform again, on demand. Once a model can reconstruct a realistic distribution of what people actually ask, the premium on having observed it collapses. The cost of doing it yourself falls at the same time. Running a prompt cluster a few hundred times and doing the statistics on it is not an expensive computation, and it gets cheaper every year. Today you buy a platform for that. Soon it is a scheduled job inside your own stack, and nobody buys a dashboard for a scheduled job. Then the last squeeze, and it is the one I would watch. The labs own the only genuinely unobservable part of this. As the model itself commoditises, and it is commoditising, they need revenue that is not tokens. Selling structured visibility into what happened between the prompt and the answer is the obvious diversification, and they are the only ones who can sell it without asking anybody's permission. On that day, every independent GEO company is reselling something its own supplier can offer directly. So expect noise. Big rounds, big titles, a category that looks like the next Semrush for about three years. Some of that money is real. Most of the moats are not. I still think it is one of the best places to be. I am less sure it is a good place to stay. I found all of this during ten days at the Hexa incubator. You show up, they hand you a problem, and your job is to prove their idea has merit. That's the deal. It isn't a thought experiment. The idea came out of those ten days with enough behind it that Hexa will put around €750,000 into building it, and is carrying it forward with my partner. I wasn't selected to keep going, but it was still one of the best ten days I've had in a long time. Real problem, real users on the phone, real market, no theatre. The work stood up. I'd do it again tomorrow (but this time with YC 🤣). --- # The Wave URL: https://www.askgib.com/articles/the-wave Published: 2026-08-10 Reading time: 4 min > A driver stops at the crosswalk. You raise your hand. They nod. The whole exchange takes a second. Most people don't do it anymore, and that quiet absence is doing something to us. You cross a street. A car stops to let you pass. The right thing to do is raise your hand. A small gesture. A nod. Maybe a smile. It takes less than a second. It costs nothing. The driver knows you saw them. You acknowledged that they did something they didn't have to do. The interaction is closed cleanly. Both of you continue your day slightly better than you would have otherwise. Now look at any street in any major city. Count how many people do this. It's gone almost completely. People walk in front of stopped cars with their faces in their phones, headphones in, no eye contact, no acknowledgment. The driver waited, the pedestrian walked, the moment passed, and nothing happened between two humans who briefly shared the same space. I want to talk about why this matters more than it seems to. For most of human history, the texture of a society was made up of micro-interactions. You walked past someone, you nodded. You walked into a shop, you said good morning. Someone held a door, you said thank you. You sat next to a stranger, you exchanged a few words about the weather. None of these were significant individually. Collectively, they were the connective tissue of a community. You knew you lived among other people because you bumped into them, lightly, all day long. The acknowledgments were small. The cumulative effect was the sense that the world contained other humans who saw you and whom you saw. | Action | Used to be | Currently | |---|---|---| | Driver stops at crosswalk | Pedestrian waves / nods | Pedestrian doesn't look up | | Someone holds a door | Verbal thank you, eye contact | Mumbled "thanks" while walking | | Entering small shop | Greeting to staff | Direct walk to product | | Bus driver opens door | Thank you exiting | Silent exit | | Stranger on transit | Brief eye contact, sometimes brief conversation | Headphones on, phone up | | Passing neighbor | Nod, hello | No acknowledgment | *Each of these used to be expected. Now each is rare enough that you notice when it happens.* Walk through a city in 2026. Most of those interactions are gone. Headphones in. Phones up. Eyes down. The texture of social life has been thinned to the point of near-disappearance. You can go through an entire day in a busy city without making real eye contact with another person who isn't paid to look at you. This isn't a small thing. This is a major shift in what living in a society *feels* like. And it's happened so gradually that we don't notice. Try this from the driver's side. You stop for a pedestrian. They walk in front of your car looking at their phone. They don't look at you. They don't wave. They don't acknowledge that you gave them the space. How do you feel? Most people, honestly, feel a small spike of resentment. Not enough to be rude. Not enough to ever say anything. But enough that the next time you see someone stepping into a crosswalk, you might not stop as readily. You might roll through. You might consider yourself a less generous driver than you used to be. This is how civic norms collapse. Not in dramatic acts of rudeness. In thousands of tiny moments where one party gives, the other party doesn't acknowledge, and the giver slowly stops giving. The pedestrian walking with their head down isn't being malicious. They probably just don't notice. But the cumulative effect of a generation of pedestrians walking with their heads down is a generation of drivers who stop noticing they should slow for pedestrians, because nothing positive ever came from doing it. The norm decays from both ends. Two things happened roughly together. The first is obvious: phones. Constant access to entertainment, communication, work, and stimulation in your pocket has made the ambient experience of being out in the world unbearable for most people. The cost of looking up from your phone went up because looking up means missing something. So nobody looks up. The second is less discussed: we stopped teaching it. There used to be an entire layer of civic education, explicit, deliberate, sometimes annoying, about how to behave in public. Hold doors. Say please and thank you. Make eye contact. Greet shopkeepers. Acknowledge people who do small things for you. This was taught at home, at school, in religious institutions, in scout troops and youth groups, and reinforced through years of social pressure. Most of that infrastructure is gone or weakened. The civic books that used to be standard in primary education in France, Britain, and elsewhere have thinned. Many schools no longer explicitly teach courtesy. Many parents are too exhausted or too distracted to enforce it. Many immigrants from regions with very different norms about public interaction never received this kind of training in the first place, and the receiving countries have failed to teach it on arrival. This isn't anyone's fault, individually. It's a quiet civic collapse that everyone is participating in. What I'm asking for is incredibly small. When a car stops for you, raise your hand. A wave, a nod, anything that says I saw you. When someone holds a door, look them in the eye and thank them, not a mumbled "thanks" at the floor. When you walk into a small shop, say hello to the person behind the counter before you start looking around, and say goodbye, or have a nice day, or au revoir on your way out. You are not transacting with a kiosk. You are dealing with a person whose entire day is being spent on people who pretend they aren't there. When you walk past a neighbor, even one you don't know, nod. Take your headphones off in shops. Make eye contact with the checkout, the bus driver, the waiter. The phone can come out later. The transaction can wait thirty seconds. If you're a parent, teach your kids. They will learn the behavior from watching you. Ten years of seeing you do it makes it natural to them. Ten years of seeing you scroll past everyone makes the opposite natural instead. If you're an educator, teach it. The civic vocabulary of public courtesy is being lost because no one is being deliberately taught it anymore. Bring it back. It's not optional culture. It's how a society holds together at the bottom level. Civilization is not held together by laws or institutions or governments. Those are the visible part. They function because the population beneath them has thousands of small, voluntary, daily interactions that affirm we are all in a shared project. When those interactions go, the institutions don't immediately collapse. They just become increasingly hard to defend. People stop feeling like they live in a society and start feeling like they live in a logistical zone with strangers. Trust falls. Politics gets more extreme. Communities thin out. Loneliness becomes endemic. Try it tomorrow. Wave when a car stops for you. Watch the driver's face. They'll be surprised. That's the saddest part. --- # The AI Moat Illusion URL: https://www.askgib.com/articles/the-ai-moat-illusion Published: 2026-07-27 Reading time: 3 min > AI models are becoming infrastructure. As intelligence becomes cheaper and more abundant, durable value will move toward products, workflows, distribution, trust and integration. *"We have no moat. And neither does OpenAI."* When Google's internal memo leaked in 2023, many people dismissed it as pessimism. Looking back, I think it was one of the most important observations made during the AI revolution. Not because Google was losing. But because it recognized something much bigger. The model itself was already beginning to commoditize. Three years later, that prediction seems increasingly difficult to ignore. After attending the Paris Open Source AI Summit and speaking with founders, researchers, infrastructure companies and chip manufacturers, I left with a very different view of where AI is heading. I no longer believe the long-term battle is about who owns the smartest model. I think we're watching intelligence itself become infrastructure. And history has shown us what eventually happens to infrastructure. Electricity was once a competitive advantage. Today it's an expectation. Owning servers was once a competitive advantage. Cloud computing turned it into a utility. Storage was expensive. Bandwidth was scarce. Compute was rare. Eventually they all became abundant. The value didn't disappear. It simply moved somewhere else. AI feels like the next chapter of the same story. Which model reasons better? Which model writes better code? Which model wins another leaderboard? Those questions matter today. I'm not convinced they'll matter nearly as much ten years from now. The remarkable thing about open source isn't simply that it's catching up. It's that it has fundamentally changed the economics of progress. Every new model teaches the next generation. Researchers build on each other's work. Optimization techniques spread across the entire ecosystem. Ideas move faster than companies. History suggests that once knowledge escapes into the open, it rarely returns behind closed doors. That doesn't mean proprietary models disappear. It means maintaining a permanent lead becomes extraordinarily difficult. The frontier keeps moving. Everyone runs after it. Eventually yesterday's frontier becomes tomorrow's baseline. We've watched this happen repeatedly across technology. There's little reason to believe AI will be the exception. Better GPUs create better AI. That is certainly true today. But today's bottlenecks are not necessarily tomorrow's bottlenecks. Software has always adapted faster than hardware. The first computers needed entire rooms. Now the phone in your pocket outperforms them. Not because physics changed. Because engineering improved. AI is following exactly the same trajectory. We're already seeing it. Quantization. Distillation. Mixture-of-Experts. Sparse architectures. KV-cache optimization. Better attention mechanisms. Specialized inference engines. Every few months someone discovers another way to achieve similar performance using fewer parameters, less memory and less energy. That trend is unlikely to stop. The opposite is far more likely. The models of 2030 will probably feel absurdly efficient compared to those of today. Not because hardware alone became dramatically better. Because we finally learned how to use it properly. This is why I don't believe the long-term winners are automatically today's hardware companies either. NVIDIA has executed extraordinarily well. AMD continues to improve. Tomorrow another company may do the same. Or perhaps the next breakthrough comes from China. Or from Europe. Or from a startup nobody has heard of yet. Five years is an eternity in semiconductors. Intel once looked untouchable. IBM looked untouchable. Cisco looked untouchable. Sun Microsystems looked untouchable. Technology has a habit of humbling incumbents. Meanwhile, the software keeps improving. Models become smaller. Inference becomes cheaper. Architectures become more efficient. Eventually, companies stop asking for the largest GPU they can afford. They start asking for the cheapest hardware capable of running their workload. Then another optimization arrives. Then another. And another. Hardware improves. Software improves even faster. The cycle repeats. It will produce models optimized for specific hardware. Apple Silicon. AMD. NVIDIA. Qualcomm. RISC-V accelerators. Custom ASICs. Entire companies will emerge whose only business is making one family of models dramatically faster on one family of chips. That is how technology evolves. General-purpose systems eventually become specialized. Specialization drives efficiency. Efficiency drives commoditization. And commoditization changes where value is created. That makes perfect sense while frontier intelligence remains scarce. But scarcity rarely survives technological progress. As models improve, become cheaper, and increasingly run locally, token pricing will come under pressure. That doesn't mean inference becomes free. Electricity isn't free. Cloud storage isn't free. Bandwidth isn't free. Infrastructure still generates enormous businesses. But infrastructure businesses rarely command monopoly economics forever. Competition compresses margins. Customers gain alternatives. Reliability, integration and service become more important than exclusivity. The same transition is likely to happen in AI. Selling raw intelligence will become less differentiated. The value will migrate toward products, workflows, trust, distribution and integration. Perhaps the biggest mistake we make is assuming someone has already found a permanent moat. History suggests otherwise. Eventually it creates abundance. Abundance changes what people value. The internet made information abundant. Search became valuable. Cloud computing made infrastructure abundant. Software became valuable. AI may make intelligence abundant. Something else will become valuable. I don't know exactly what that "something else" is. Perhaps it's distribution. Perhaps it's trust. Perhaps it's owning the customer relationship. Perhaps it's something none of us has imagined yet. What I am increasingly convinced of is this: The model is not the destination. It is becoming the foundation. And foundations rarely capture the greatest share of value forever. Five years from now, we may look back and laugh that we once believed intelligence itself could be a moat. History rarely rewards people for owning yesterday's bottleneck. It rewards those who recognize where value is moving next. --- # Do I Still Got It? URL: https://www.askgib.com/articles/do-i-still-got-it Published: 2026-07-20 Reading time: 2 min > After years of entrepreneurship without a scoreboard, I entered the first three hackathons of my life to find out whether I could still figure things out. A few weeks ago, I wanted to answer a simple question. **Do I still got it like that?** Over the course of twelve days, I entered the first three hackathons of my life. One organized by Y Combinator. One at Google. One with Anthropic. Three dubs. The trophies were satisfying, but they were never the objective. The objective was the test. There are no annual reviews. No promotion cycles. No rankings. No one tells you whether you're improving or slowly becoming less relevant. You spend months, sometimes years, working on a single company, solving the same problems, speaking to the same people and living inside the same world. Eventually, it becomes difficult to measure yourself honestly. Your friends encourage you. Your team believes in you. Customers give you mixed signals. Investors have their own incentives. Everyone has an opinion. Very little of it is objective. I wanted to walk into a room where nobody knew me. Nobody cared what I had built before. Nobody owed me anything. Just a clock. A problem. A team. And qualified judges who would evaluate only what happened over the following twenty-four or forty-eight hours. There is something refreshing about that level of honesty. There is no time for endless planning. No opportunity to polish a presentation for months. No carefully constructed narrative explaining why success is just around the corner. You either identify the right problem or you don't. You either build something meaningful or you don't. You either communicate your vision effectively or you don't. Everything happens under pressure. That pressure reveals interesting things. Can you make decisions with incomplete information? Can you bring strangers together into an effective team? Can you recognize which ideas deserve attention and which should be abandoned? Can you remain calm while the deadline approaches? (low cortisol gang) Those questions matter far more than whether you remember the syntax of a particular programming language. Winning those three competitions didn't convince me that I'm somehow extraordinary. Three hackathons prove very little in the grand scheme of a career. What they gave me was something much smaller, and much more valuable. Not confidence that I will always succeed. Confidence that I can still walk into uncertainty and create something meaningful. Entrepreneurship slowly erodes that certainty. When projects take years, it's easy to confuse slow progress with personal decline. You begin asking uncomfortable questions. Have I lost my instincts? Am I becoming slower? Was previous success simply good timing? Am I still capable of building something people genuinely value? Those questions probably never disappear. Nor should they. But every now and then, it's healthy to expose yourself to an environment that provides immediate feedback. Hackathons do exactly that. They force you to think clearly. To prioritize. To execute. To communicate. They remind you that entrepreneurship isn't about knowing every answer before you begin. It's about becoming comfortable solving problems you didn't expect to encounter. I wasn't testing whether I could still build software. I wasn't testing whether I could still pitch. I wasn't even testing whether I could win. I was testing whether I could still figure things out. Because that's the real job of an entrepreneur. Not predicting the future. Not having perfect plans. Not always being the smartest person in the room. The real job is entering situations filled with uncertainty and finding a path forward anyway. Sometimes that takes years. Sometimes it takes forty-eight hours. Either way, the skill is the same. That's what those three weekends reminded me. Sometimes, the best way to measure yourself isn't by looking backward at what you've already built. It's by putting yourself somewhere completely unfamiliar, giving yourself an impossible deadline, and seeing what happens next. --- # The Hackathon Economy: Why AI Changed the Game URL: https://www.askgib.com/articles/the-hackathon-economy Published: 2026-07-13 Reading time: 3 min > AI has not just changed how we build software. It has changed who gets to build it, and hackathons make that transformation impossible to miss. Over the past year, I've become convinced that AI hasn't just changed how we build software. It has changed who gets to build it. Hackathons are perhaps the clearest place to see that transformation. Most people still imagine hackathons as coding competitions. They picture rooms full of software engineers racing to write the best algorithm before the clock runs out. That isn't what they are anymore. Twenty years ago, the person who could write the fastest code or optimize the cleverest algorithm often had a decisive advantage. Today, large language models have dramatically lowered that barrier. Code has become increasingly abundant. Ideas have not. The scarce resources are now judgment, creativity and execution. The winning team is rarely the one that writes the most code. More often, it's the team that identifies the right problem, builds just enough to prove the idea, and explains why the solution deserves to exist. That is a very different competition. Every judge has already seen hundreds of AI demos. Connecting five APIs together or orchestrating a chain of agents is no longer impressive by itself. Those have become commodities. **Who actually needs this?** If the answer is nobody, you've spent an entire weekend building something that will quietly die in a GitHub repository. The biggest shift of the post-AI era is that building has become cheap. Thinking has become expensive. For the first time, someone doesn't need to have spent fifteen years becoming an expert software engineer to build something meaningful. Founders, designers, doctors, farmers, lawyers, researchers and operators all have a genuine opportunity to compete because they often understand the problem better than anyone else in the room. They may never become the strongest programmer. They don't have to. Technology has become increasingly democratized. Domain expertise has become increasingly valuable. Ironically, AI has made software less about software. It has made it more about understanding people. The result is that hackathons have become more competitive than ever before. You're no longer competing only against programmers. You're competing against people who have lived the problems they're solving. Against founders who understand markets. Against designers who understand users. Against researchers who understand science. Against operators who understand how businesses actually function. The playing field has expanded, and the level of competition has risen with it. Today's hackathons often include hundreds of thousands of dollars in cash, cloud credits, GPU credits, AI credits, hardware, investor introductions and hiring opportunities. For some people, the ecosystem has become substantial enough that they effectively move from hackathon to hackathon. Whenever meaningful rewards exist, competition intensifies. Some teams arrive exceptionally prepared. They understand the technology, the judging criteria and how to tell a compelling story in only a few minutes. Naturally, emotions run high. I've seen people become genuinely angry about judging decisions. They convince themselves another team didn't deserve to win or that the judges misunderstood their project. I understand disappointment. Competition should matter. You should care. You should analyze why you lost and what you could have done differently. But I've never understood complaining about a free opportunity. You were given food. You were given access to frontier AI models. You were surrounded by talented engineers, founders and investors. You built something that didn't exist when you walked through the door. You probably learned more in forty-eight hours than you would have during several ordinary weeks. Even if you leave without a trophy, you almost never leave empty-handed. That alone makes the experience worthwhile. Not perfect meritocracy. Life isn't perfectly fair, and neither are hackathons. Some people have stronger teammates. Some arrive with more experience. Some happen to build exactly what resonates with a particular panel of judges. And then there is luck. People often resist acknowledging how much luck influences success. I've never understood why. Acknowledging luck does not diminish hard work. It elevates humility. Preparation determines whether you're ready when opportunity appears. Execution determines whether you remain in contention. Luck sometimes decides which of two exceptional teams finishes first. That isn't unique to hackathons. That's life. They compress entrepreneurship into forty-eight hours. You begin with uncertainty. You build under pressure. You present before strangers. You receive an outcome you cannot fully predict. And then you begin again. AI has fundamentally changed what these competitions measure. They are no longer searching for the fastest programmer. They're searching for the best problem-solvers. Those are very rarely the same people. --- # Bring Back Direct Democracy URL: https://www.askgib.com/articles/bring-back-direct-democracy Published: 2026-07-06 Reading time: 8 min > The Greeks did it. The Swiss still do it. Modern technology makes it scalable. So why do we still let 500 people decide for 70 million? Once every four years, you vote. You pick a name from a short list of people who almost certainly don't live in your city, don't know your neighborhood, and won't take your call. They go to a capital. They vote on hundreds of laws. They vote on wars. They vote on budgets in the billions. They vote on bills written by lobbyists you've never met for industries you don't work in. You agree with maybe 30% of what they do. You disagree with maybe 40%. You don't even know about the rest. This is what we call democracy in 2026. It is, in functional terms, four-year elective autocracy with a marketing department. | Layer | Approximate count | |---|---| | Voters | ~50,000,000 | | Elected representatives | ~600 | | President / PM | 1 | | Senior advisors | ~10 | *Modern "democracy" compresses millions of voters into a handful of decision-makers. The compression ratio is roughly 5 million to 1.* The original Greek model was direct. Every free citizen could attend the Assembly in Athens and voted personally on every major decision, war, peace, public expenditure, exile of officials. The Council of 500, which set the agenda, was selected by lot from the citizenry. Members rotated. No one accumulated permanent power. Decisions were executed by people who would go back to being farmers and craftsmen a year later. But be honest about what Athens was. It was small. About 30,000 to 60,000 men held citizenship, in a region of maybe 250,000 to 300,000 people. Women, slaves and foreign residents had no vote. From 451 BC you needed two Athenian parents to be a citizen at all. So this was a small club of men who looked alike, prayed alike and could all walk to the same hill. They ran it with no printing press and no fast way to send a message. And it broke, more than once. Before the democracy, one family of tyrants ruled the city for most of 50 years. In 411 BC a group of 400 rich men took over in a coup. In 404 BC, after Athens lost its long war with Sparta, thirty men ruled by terror and killed about 1,500 citizens in less than a year. The crowd made its own disasters too. In 427 BC the Assembly voted to kill every man in the rebel city of Mytilene, then voted again the next day to take it back. In 406 BC it put six of its own winning generals to death. In 399 BC a jury of citizens killed Socrates. The structural principle was still sound. Citizens did not hand their power to a permanent class, and each time someone took it, they took it back. But Athens is proof that the idea can work. It is not a design you can copy. The Romans had a parallel structure. The Senate held aristocratic power, but the *tribunes of the plebs* could veto Senate decisions, propose laws directly to the people, and held the legal right of *sacrosanctitas*, they couldn't be harmed, and harming one was a capital offense. Consuls served one year only. Power was distributed, contested, time-limited. On paper. In practice the men at the top kept inventing ways around it. They bought votes, and Rome kept passing new laws against bribery because the old ones never worked. They turned the veto into a weapon. In 133 BC the Senate's allies got one tribune to veto another, and that same year senators beat the reformer Tiberius Gracchus to death. In 59 BC a consul tried to cancel a whole year of laws by saying he was watching the sky for bad omens. The result was stalemate. Land and debt reform was fought for a century by the *optimates*, the Senate's conservative side. Stalemate turned into blood. Sulla marched an army on Rome in 88 BC. Caesar did it again in 49 BC, after the optimates ordered him to give up his army. The Republic lasted nearly five centuries, and its last one was civil war after civil war. It did not fall because the people had too much power. It fell because a small class learned to block everything, until the only man who could get anything done was a man with an army. That is the Empire, which is the form we somehow ended up copying. And then there's Switzerland, which has kept direct democracy alive into the modern era. In the cantons of Glarus and Appenzell Innerrhoden, citizens still gather in an open-air assembly, the *landsgemeinde*, and vote on laws by show of hands in the public square. At the federal level, Swiss citizens vote four times a year on referendums covering taxes, immigration, foreign treaties, constitutional changes, almost anything significant. Any citizen group can force a referendum on a new federal law by collecting 50,000 signatures in 100 days. The result: the Swiss government, more than any in the developed world, actually has to make policy that the population agrees with. We took the bad part of the Roman model, concentrated power in a small assembly of permanent professionals, and dropped the good part, the citizen veto. You elect a representative. From the moment they take office, their incentives are no longer aligned with yours. They answer to: The party machinery that controls their nomination for the next election. The donors who fund their campaign. The lobbyists who write the bills and explain them to staff. The career officials in their department. The leadership of their chamber, which controls committee assignments. You are item number six on this list, at best. The system isn't broken in the sense that it doesn't work as designed. It's broken in that what it was designed for stopped serving the people about a century ago. *Every delegate is one step removed from the people who chose them. A recall vote opens when 5% of them ask for it. No one accumulates permanent power.* Citizens organize at the local level. Every 1,000 citizens elect one delegate from among themselves. The delegate isn't a politician, they're a neighbor with a job who took on a civic responsibility for a fixed term. The 1,000 delegates from a region elect, among themselves, 100 delegates at the regional level. Same recall rules. Same time limits. The 100 regional delegates from each region elect 10 at the national level. The 10 advise on national policy. There is no permanent political class. There are no donors. There are no parties, or at most, loose affiliations. Every delegate is accountable directly to the people who chose them, and **recall is always one step away.** If a small share of the people who chose you, say 5%, sign against you, a recall vote opens. If most of them vote you out, you are removed. No election cycle. No waiting four years. Without recall, the system reverts to ordinary representation, which is what we have and which doesn't work. With recall, the incentive flips: delegates must continuously represent the actual views of their electors, because those electors can pull the rug at any moment. **Liquid democracy** is the academic name for a system where citizens can vote directly on any issue, *or* delegate their vote on a specific topic to someone they trust, *and* revoke that delegation at any time. The German Pirate Party was the first to run a working platform for this, LiquidFeedback, and used it for internal governance. Loomio in New Zealand built smaller-scale versions. Estonia has experimented with parts of this through its digital infrastructure. **Delegative democracy** is a slightly different model where delegates carry the actual votes of the people who chose them, so a delegate representing 1,000 citizens casts 1,000 votes, not one. This makes accountability visible: you can see exactly how they voted on every issue, and whether they voted the way their electors wanted. Both models address the central problem of representative democracy: representation, once given, can't be taken back fast enough to matter. The four-year cycle is too slow. The modern world produces important decisions every week. This idea was attempted in Libya from 1977 onward under Muammar Gaddafi. His *Green Book* laid out a similar pyramidal structure of People's Congresses and People's Committees, with delegates revocable from below. Citizens were automatic members of their local congress. There were no political parties. It didn't work, and the failure is instructive. Gaddafi maintained a parallel "Revolutionary Command" structure outside the formal democratic apparatus, controlling the military, security services, and oil revenue. The People's Congresses were real, but had no power over what actually mattered. The theory was sound. The implementation was hijacked. The lesson isn't that direct democracy doesn't work. It's that any version of it has to be designed with the assumption that someone will try to hollow it out from above. The Swiss model survives because there is no shadow Revolutionary Command. The Athenian model came back from two coups because its citizens would not let any one man become permanent. The formal structure isn't enough on its own. A representative votes for a war. They don't fight. Their children don't fight. They go home to a security detail. A representative votes for a tax cut funded by debt. They retire on a pension before the bill comes due. A representative votes for an environmental rollback. They live in a wealthy neighborhood unaffected by the pollution. When the people making decisions have to live with the decisions, the decisions get better. Not perfect. Not always wise. But aligned, in a way that current representation simply isn't. A neighbor who has to face the people they represent every week at the grocery store is going to vote very differently from a national legislator who flies home once a month. The accountability loop, currently broken, gets reconnected. Real cost analysis. This isn't free. Direct democracy is slow on most issues. Foreign crises sometimes need a decision in twenty minutes, and consulting a thousand delegates can't deliver that. Some functions, defense in wartime, central bank operations, fast-moving regulatory responses, genuinely require small groups acting fast. The fix is hybrid. Most decisions go through the pyramid. A defined small list, emergency response, certain technical regulation, monetary policy, stays with smaller bodies with sunset clauses and post-hoc citizen review. Most of the important decisions still go to the thousand. The oldest objection is the strongest one. Give the crowd direct power and you get demagogues, then dictators. The American founders argued about exactly this. In 1776 Pennsylvania wrote the most radical constitution of the new states: one chamber, elected every year, no governor with a veto, and every public bill printed for the people and held until the next session, except in an emergency. Many founders hated it. James Madison wrote that pure democracies "have in general been as short in their lives as they have been violent in their deaths." Pennsylvania dropped its design in 1790. The United States got a Senate, a president and a four-year wait. They were right about the risk and wrong about the cure. Athens shows the risk: one good speaker and one angry afternoon. But the cure they chose was distance between the people and power, and that distance is the thing that rotted. The pyramid uses a different cure. Nobody speaks to a crowd of millions. A delegate speaks to people who know them. Big decisions get a second vote after a delay, the way Athens saved Mytilene and Pennsylvania made its bills wait. There's also the risk of populist whiplash. Direct democracy in the wrong configuration produces decisions like Brexit, popular votes on enormously complex issues without the institutional weight to think them through. The Swiss avoid this with a culture of small, frequent referendums that build civic literacy over time. They ask about ten national questions every year, and many more local ones, and the population gets practiced. The current system is designed badly *and* doesn't represent the population. A direct system designed thoughtfully wouldn't be perfect, but it would close the gap between what people want and what their governments do. We are running 21st-century societies on 18th-century governance technology. The American Constitution was written when getting information from a far state capital to the national capital took two weeks on horseback. Of course we delegated everything to a small group of professionals, there was no other choice. That constraint is gone. Every citizen has more communication infrastructure in their pocket than the entire United States government had in 1850. Decisions that used to require months of physical travel and printed pamphlets can now be made by millions in an afternoon, with deliberation, accurate information, and traceable accountability. The technology is here. The theoretical models are here. The historical precedents, Athens, Rome, Switzerland, are here. What's missing is the political will to build it, because the people who currently hold power are exactly the people who would lose it. Direct democracy is not a fantasy. It's a question of whether we want to keep paying a permanent political class to make our decisions for us, or whether we want to start making them ourselves. The Greeks figured this out 2,500 years ago. We can probably manage it too. --- # France's Electricity Suicide URL: https://www.askgib.com/articles/france-electricity-suicide Published: 2026-06-29 Reading time: 3 min > France makes some of the cheapest electricity on Earth and bills its people some of the most expensive in Europe. The sequel to the tax piece: same disease, different organ. France makes some of the cheapest electricity on Earth. French people pay some of the most expensive bills in Europe. Read that again. *(If you read my piece on French taxes, this is the sequel. Same disease, different organ.)* Start with the fact nobody disputes. France runs the biggest nuclear fleet in Europe. 57 reactors. State-owned. Cost to produce a kilowatt-hour: about 6 cents. Among the cheapest, cleanest power on the planet. So why is your bill brutal? Two reasons. Both self-inflicted. Europe has one rule for pricing power. The most expensive plant running sets the price for everyone. Nuclear runs first. It's cheap. Doesn't matter. The last plant to switch on is usually gas. That sets the price. And every producer gets paid that price. So France makes power at 6 cents and bills you like it burned Russian gas to make it. The gap doesn't vanish. EDF keeps it. You pay gas prices for nuclear electricity. In the most nuclear country in the world. It got worse. For years France was forced to sell its cheap nuclear to competitors at 42 euros. A law called ARENH. 2022, peak madness: middlemen bought EDF's power at 46 and sold it back to EDF at 257. EDF lost 17.9 billion euros in one year. You can't write satire this good. And here's the part that should make you put down your coffee. It costs about 6 cents to make a French kilowatt-hour. You pay around 22. The electricity is the cheapest thing on your electricity bill. The rest is grid, taxes, and margin. Everything stapled to the power costs more than the power. There's a scheme called CEE. Energy-saving certificates. The state forces suppliers to fund insulation, new boilers, "efficiency." Sounds nice. Here's the catch. The state doesn't pay for it. It's bolted onto your bill. Around 10 billion euros a year. Opaque. Riddled with fraud. Fake renovations, fake paperwork. Think about the logic. You have the cheapest power source in Europe sitting right there. And the grand plan is to spend billions begging you to switch off the lights. You don't ration water next to a river. Germany's electricity is expensive. Gas-dependent. A mess. So what does Germany do? Subsidizes its factories down to 5 cents a kilowatt-hour. Pays them to stay. France has the 6-cent power and subsidizes nobody. So the factory moves to Germany. Buys subsidized power. Sometimes French power, flowing across the border. Builds the product. Ships it back. And the French buy it. At a markup. We export the cheap electricity and import the finished good. Paying extra to use our own advantage against ourselves. France already figured out that cheap nuclear is bait. It just refuses to share it with you. SoftBank put 75 billion euros on the table for AI data centers in France. Why? Energy is half the cost of running AI, and France's grid is the cheapest clean power in Europe. 109 billion got pledged at a single summit. Amazon, Microsoft, the Emiratis, all of them. The global AI buildout is a 7 trillion dollar wave this decade. It runs on one thing: cheap, clean power. France has it in a quantity almost no one else does. So France knows exactly what its electricity is worth. It sells that price to Microsoft. It bills you 22 cents. One catch, and it's important. Most of that "investment" is pledges. Announcements, not poured concrete. France keeps winning the press release. Whether it wins the factory is another question. And the gap shows. France's flagship lab, Mistral, has raised about 4 billion dollars. OpenAI has raised 186 billion. Anthropic, 161 billion. The bottleneck was never electricity. France is one of the biggest net payers into the EU. Around 26 billion euros a year out the door, and about 9 billion of it never comes back. A lot of it flows east. To countries that then offer cheaper labor, lighter charges, and pull the factories that should've been ours. We fund our own competitors. Then buy back what they build. Sell electricity to French homes and industry at cost. Strip the windfall and the tax pile and you're near 12 cents tomorrow. Then take EDF's profits and pour them into the grid and new reactors. The price keeps falling. 10 cents. 9. 8. The advantage compounds. Every year it gets cheaper, and every year it gets more insane to leave it on the table. Suddenly France is the cheapest place in Europe to run anything. Factories come here. Data centers come here. Jobs come here. This is the cleanest argument for renegotiating France's place in the EU energy market that exists. At minimum the deal gets reopened. At maximum it's the first honest case for leaving. Not emotion. Not flags. Just math. France has no winning move in the current setup. It's pure loss, top to bottom. And it poisons everything, because energy is the thing every other thing needs to work. France was handed the best energy hand in Europe. And it's playing it face-down, on purpose. --- # The Real French Tax Rate URL: https://www.askgib.com/articles/the-real-french-tax-rate Published: 2026-06-22 Reading time: 2 min > Put €1,000,000 in one person's pocket and watch what's left. Salary, you keep 26%. Dividends, 44%. The real French tax rate isn't a number, it's a set of incentives. Say a company has €1,000,000 and wants to put it in one person's pocket. Watch what is left by the end. First, kill a myth. People think the company pays 25 percent corporate tax no matter what. It doesn't. Salary is a deductible expense, so if you pay the whole million as salary, the company shows no profit and owes almost no corporate tax. The 25 percent only hits money you keep. So there are two doors, and they do not cost the same. Door one is paying it as a salary. The employer's charges eat about €300,000 before the employee sees a cent. The employee's own charges take about €155,000. Income tax, top bracket at 45 percent plus a surtax, takes about €230,000. Then they go to spend what is left, and VAT takes a fifth of that. What actually reaches a human being is about €262,000. The state took three quarters of it. Door two is keeping it as profit and paying a dividend. Now the 25 percent corporate tax finally shows up, €250,000 gone. The €750,000 left goes out as a dividend, taxed at a flat 30 percent. Then VAT on what is spent. What reaches the person is about €437,000. Read those two numbers again. Salary, you keep 26 percent. Dividend, you keep 44 percent. Working for your money is taxed harder than owning it, almost twice as hard. So what does any rational owner do. Pay everyone, including himself, the minimum. Show as little profit as the law allows. Take the rest as dividends. This is why so much of France looks the same, small companies that somehow never turn a profit, everyone on a modest declared salary, a system that quietly rewards you for not growing, not hiring well, not paying real wages. The tax code didn't ask for hard work. It asked for accountants. And the bigger you are, the more doors you have. A small shop is stuck, its profit sits in France and gets taxed. A multinational plays geography. It splits into dozens of companies across dozens of countries, then sets the prices between its own subsidiaries. The French arm buys high and sells low to its sister in a cheaper country, so the profit appears in Geneva, or Luxembourg, and France is left with a company that somehow earns nothing. TotalEnergies is the clean example. Its oil flows route through Totsa, its trading arm in Geneva, a hub of nearly 1,100 people. The result, no French corporate income tax in 2020, none in 2021, a record-profit year. It happened again last year, and the CEO pointed to a loss on refining. All legal. The internal price just decides which country gets to tax you. Geneva and Luxembourg are very understanding about it. Your bakery is not invited. Now look down, at the floor. France makes the bottom cheap to hire. At minimum wage the state hands employers a break that wipes out most of their charges, so hiring at the SMIC costs barely 10 percent on top. Push a salary up toward a real wage and that break vanishes, and the charge snaps back to about 42 percent. Drag the slider and watch the green shrink as the salary climbs. So the whole structure points one way. Hire many at the floor, where it is subsidized. Pay almost no one well, because that is where it is punished. Keep declared profit near zero, or ship it abroad. The system didn't ask anyone to build, grow, or pay people properly. It asked them to optimize, so they did. That is the real French tax rate. Not a number, a set of incentives. Cheap to hire poor, expensive to pay well, free to be huge. --- # AI Just Got a Passport URL: https://www.askgib.com/articles/ai-just-got-a-passport Published: 2026-06-15 Reading time: 4 min > This week the best AI on Earth got a border. Not a paywall. A border. The most powerful tools we have ever built now check your nationality at the door. This week the best AI on Earth got a border. Not a paywall. A *border.* You need the right birth certificate now. Here's what happened. The U.S. government told Anthropic: the frontier models, the Mythos-class ones, Fable 5 and Mythos 5, are off-limits to foreign nationals. National security. A jailbreak they couldn't contain. So overnight, the most powerful tools humanity has built got a wall around them. By nationality. Now sit with the absurdity. The people who *build* these models, researchers born in Canada, the UK, wherever, might not be allowed to use the thing they made. A co-founder of Anthropic, born outside the U.S. Their own researchers, born in other countries. The company exists *because* of them. And now they might have to flash a passport to touch the thing they built. You can write the brain. You're just not American enough to be in the room with it. That's the "security risk." The person who wrote the mind is too dangerous to use the mind. Because of where his mother gave birth. Read that back and tell me it makes sense. OpenAI, it's literally in the name, started as a non-profit. Open. For everyone. If we ever cracked intelligence, it was supposed to belong to all of us. That was the pitch. The people who built Anthropic came out of that house. And to be straight, they didn't leave chasing some non-profit dream. They left over a clash of vision. They thought OpenAI was shipping too fast and not taking safety seriously enough. So they built their own lab. For-profit from day one. No "open" in the name, no pretense. But here's what nobody did. Stay open. The non-profit became a juggernaut. The breakaway became a kingmaker with a flag on the gate. Whatever the founding story was, open, or safe, or for-everyone, this isn't it. And the rest of us are tenants. AI stopped being software. It became a **resource.** Oil. Lithium. A port. And resources don't stay neutral. Countries hoard them. Fence them. Eventually fight over them. We've seen this movie. It's old. If you weren't born in the right empire, you got locked out of the last few centuries by default. Pre-colonial logic. New coat of paint. Same gate, same guards, same people standing outside it. So here's where I land. Either AI goes actually open, same access for everyone, everywhere, or it becomes the thing nations knife each other over for the next fifty years. There's no soft middle. Something this powerful, owned by one flag, doesn't stay one flag's for long. France still has a shot. It has some of the pieces, and more importantly, it has the *appetite.* The desire to stay relevant in a century that's trying to leave it behind. And I think it should be willing to sacrifice for it. Because this is no longer a political endeavor. This is a **survival** endeavor. Survival. Sovereignty. Relevance. For forty years the pattern has been the same. The country that wins the technology wins, full stop. Resources stopped being the lever a long time ago. Compute is the new oil, and we don't pump it here yet. So if I ran France, I'd make the offer of the decade. **Anthropic, move your headquarters here.** Then actually make it survivable. France taxes ambition into the ground. You can't pay frontier-lab salaries through these social charges and live. So stop forcing the future through laws written for the past. Write a new category. *Innovative infrastructure.* Different rules. Build the law for the thing you're trying to attract, not the thing you're trying to bleed. **ASML.** Dutch. European. The only company on Earth that makes the machines that make the most advanced chips. The entire planet's silicon runs through one building on this continent. America wants to wall off models? Europe holds the wall that makes the wafers. That's not a small card. That's the whole table. And the Nvidia dependency is on a timer anyway. China is already training on its own chips. The "you need American GPUs" era has an expiry date. Betting the next decade on Nvidia staying king is a bad bet. Yeah, it costs. A few hundred billion, cash, to build it. Same again every year to grow it. That's the price of not renting someone else's century. Frontier AI doesn't run on code. It runs on electricity. Oceans of it. The biggest line item in a data center isn't the chips. It's the power bill that never stops. France has the cheapest clean power on the continent. Fifty-seven reactors, already built, already paid for. The capital, the construction, the decades, that bill was settled years ago. What's left is fuel and upkeep. Under a cent a kilowatt-hour for the fuel of the next unit, and even all-in, run at cost with no profit, under eight. The plants are sunk cost. The power coming out of them is nearly free to make. And right now France sells that power into the European market at European prices. It exports its own advantage. It hands the discount to everyone else. So flip it. Stop shipping hundreds of billions abroad, to other people's wars and other people's recovery. Spend it here. Build new reactors aimed at one job, feeding compute. Point the cheapest electricity on Earth at the most valuable machines on Earth, on the same soil. It's not a subsidy. It's a moat nobody else has. If everyone holds the best model, no one can be hit with it. Equal footing. Mutual defense by default. The danger was never powerful AI. The danger is powerful AI that only *some* people are allowed to hold. This week we learned that intelligence, the most human thing there is, can be issued by nationality. That should scare you more than the machine ever did. --- # The Anti-Nutrients Nobody Tells You About URL: https://www.askgib.com/articles/the-anti-nutrients-nobody-tells-you-about Published: 2026-06-08 Reading time: 6 min > Spinach has 1,145mg of oxalates per 100g. The oxalates bind to the iron and calcium in the spinach. The label tells you about the iron. It doesn't tell you that you're absorbing almost none of it. Spinach is famous for being full of iron and calcium. Popeye built his career on it. Generations of children have been told to eat their greens because greens are loaded with the minerals your body needs. Here's the part the label doesn't tell you. Spinach contains an average of **1,145 mg of oxalates per 100 grams**. Oxalates bind to calcium, iron, magnesium, and zinc, the same minerals spinach is supposed to provide. By the time your body has finished processing a serving of spinach, you've absorbed only a fraction of the iron and almost none of the calcium that the nutrition label promised. The label tells you what's in the food. It doesn't tell you what your body can actually use. Plants don't want to be eaten. This is the central fact of plant nutrition that we somehow forgot in popular discussions of food. Animals can run away from things that want to eat them. Plants can't. So plants have evolved a chemical arsenal of compounds that make them harder to digest, less nutritious to consume, or actively unpleasant to eat. These compounds are designed to discourage being eaten, and they often work by blocking the absorption of the nutrients in the very plant that contains them. These compounds are called **anti-nutrients**. They are not poisons. They're not going to kill you in normal dietary amounts. But they significantly reduce how much of the food's listed nutrients you actually absorb. And the nutrition label, by industry convention, only reports what's *in* the food, not what's *bioavailable*. | Anti-nutrient | Found in | What it blocks | |---|---|---| | Oxalates | Spinach, chard, beets, almonds, tea | Calcium, iron, magnesium, zinc | | Phytates | Whole grains, legumes, nuts, seeds | Iron, zinc, magnesium, calcium | | Lectins | Beans, peanuts, soy, whole grains | Calcium, iron, phosphorus, zinc | | Goitrogens | Broccoli, kale, cabbage, brussels sprouts | Iodine (affects thyroid) | | Tannins | Tea, coffee, wine, legumes | Iron | | Saponins | Legumes, whole grains | General mineral absorption | *You're eating these compounds every day. The label tells you about the iron and the calcium. It doesn't tell you that a substantial portion is being blocked from absorption by other things in the same food.* **Oxalates** are found in spinach, Swiss chard, beets, rhubarb, almonds, sweet potatoes, and tea. They bind to calcium and iron in the gut, preventing absorption. In people prone to kidney stones, they form calcium oxalate crystals, the most common type of kidney stone. **Phytates** (phytic acid) are concentrated in the bran of whole grains, in nuts, in seeds, and in legumes. They bind to iron, zinc, magnesium, and calcium. The whole-grain bread that's supposedly full of minerals is also full of phytates that block those minerals from absorption. **Lectins** are found in legumes (beans, peanuts, soybeans) and whole grains. They can interfere with the absorption of calcium, iron, phosphorus, and zinc. Some lectins also irritate the intestinal lining when consumed in large quantities; this is why undercooked kidney beans can cause severe digestive distress. **Goitrogens** are found in cruciferous vegetables, broccoli, kale, cabbage, Brussels sprouts, cauliflower, mustard greens. They can prevent the absorption of iodine, which is required by the thyroid. In people with iodine deficiency or hypothyroidism, high consumption of raw cruciferous vegetables can worsen thyroid function. **Tannins** are in tea, coffee, wine, and legumes. They decrease iron absorption. The standard advice not to drink tea or coffee with iron-containing meals comes from this. **Saponins** are in legumes and whole grains. They can interfere with normal nutrient absorption. | Nutrient (per 100g raw spinach) | Listed | Bioavailable | |---|---|---| | Calcium | 99 mg | **~5 mg (5%)** | | Iron | 2.7 mg | **~0.05 mg (2%)** | | Magnesium | 79 mg | reduced significantly | | Oxalates (anti-nutrient) | 1,145 mg total / 803 mg soluble | binds the minerals above | *Source: Petroski & Minich, 2020; Heaney et al. 1988 on calcium absorption from spinach. The label tells you what's in the food. It doesn't tell you what your body can use. For spinach, the gap is roughly 10x.* The oxalates outweigh the minerals they're going to bind to by a factor of roughly **11 to 1**. This is not a tie. The oxalates win. When researchers actually measure how much calcium from spinach gets absorbed, the answer is **about 5%**. So out of 99 mg of calcium in the food, you absorb about 5 mg. For comparison, calcium absorption from milk is roughly **30%**. To get the same usable calcium as one glass of milk, you'd need to eat several pounds of spinach. The iron is slightly less affected, because heme iron and non-heme iron behave differently and because vitamin C in spinach helps with absorption. But you're still absorbing a small fraction of what the label suggests. This doesn't make spinach bad. Spinach has other things in it, folate, vitamin K, fiber, antioxidants, that are valuable independent of the minerals. But the basic story we tell about spinach, that it's an excellent source of iron and calcium, is at best half-true. The same pattern repeats across almost every plant food on the planet. Try a few, search by name or category: | Method | Anti-nutrient reduction | |---|---| | Boiling spinach (12 min) | **87% oxalate reduction** | | Boiling Swiss chard (12 min) | 85% oxalate reduction | | Soaking grains overnight | 17–28% phytate reduction | | Sprouting legumes | 60%+ phytate reduction | | Fermenting (sourdough, sauerkraut, tempeh) | Significant phytate + lectin reduction | | Cooking beans thoroughly | Lectins deactivated | *Source: Chai & Liebman 2005 (oxalates); Lestienne et al. 2005; Duhan et al. 2002 (phytates); various. Traditional food preparation methods existed for a reason. The "raw is best" movement of the last 20 years has been moving in the wrong direction for mineral absorption.* **Boiling spinach for 12 minutes reduces oxalate content by 87%.** Swiss chard, similarly, drops 85%. This is one of the few cases where the traditional folk wisdom about cooking vegetables actually has a clear nutritional rationale. **Soaking grains** reduces phytate content by 17-28%. This is why traditional preparation of grains, overnight soaking before cooking, wasn't just for taste. **Sprouting legumes** reduces phytate content by over 60%. This is why sprouted lentils, mung beans, and similar are easier to digest and more nutritionally available than their unsprouted versions. **Fermenting** dramatically reduces both phytates and lectins. Sourdough bread has substantially fewer phytates than regular yeast bread. Tempeh has fewer than unfermented soybeans. Sauerkraut has fewer than raw cabbage. **Cooking** in general deactivates lectins. Raw kidney beans contain enough lectins to cause severe poisoning if eaten in quantity; properly cooked kidney beans contain almost none. The pattern is consistent: traditional preparation methods, soaking, sprouting, fermenting, prolonged cooking, that humans have used for thousands of years are usually doing meaningful anti-nutrient reduction. The "raw is best" food trend of the last twenty years has, in many cases, been moving in the wrong direction. The nutrition labels you see on packaged food are mandated by regulators. The format is standardized in most countries: calories, fats, carbs, protein, key vitamins, key minerals. In the US, the format is set by the FDA. In the EU, by a 2011 regulation. Each format went through years of regulatory negotiation. None of them include bioavailability adjustments. None of them mention anti-nutrients. None of them tell you the ratio of mineral-blocking compounds to minerals. This is an obvious gap, and it's a gap with consequences. People making dietary choices based on labels are getting a partial picture. If you're trying to increase your iron intake by eating spinach, you're going to absorb far less iron than the label suggests. If you're choosing whole-grain bread for the magnesium, the phytates are taking back most of it. If you're loading up on raw kale smoothies for the calcium, the goitrogens and oxalates are doing real work against you. A better label format would show two columns: **listed nutrients** and **bioavailable nutrients**. The first column would be the current information. The second would adjust for the anti-nutrients in the food and the cooking method indicated. For raw spinach, the bioavailable calcium would be ~5 mg, not 99 mg. For boiled spinach, more like 50 mg. The consumer would actually know what they're absorbing. The pushback to this would come from the food industry, which currently markets products on the listed nutrient content. A breakfast cereal "high in iron" might be much less impressive once the phytates are accounted for. A leafy green "rich in calcium" might be in a different class of food entirely. Companies that have built brands on the optimistic version of the label would lose marketing leverage. This is exactly why the change won't happen voluntarily. It needs to happen through regulation, the way the original nutrition labels did. And it should happen soon, because the gap between what people think they're eating and what their bodies are actually using is large and consequential. Food labels were a major regulatory achievement of the 20th century. They told consumers what was in their food. They held producers accountable. They are also out of date. They report what's in the food. They don't report what your body can use. For most foods, the gap is small enough not to matter. For high-anti-nutrient foods, the gap is enormous, sometimes the bioavailable nutrient content is one-fifth to one-tenth of what's listed. A serving of raw spinach is a celebration of plant chemistry. Most of the iron and calcium are unavailable. The labels don't say this. The nutrition guides don't say this. The cookbooks don't say this. The general cultural narrative around eating leafy greens for minerals doesn't say this. It's not a conspiracy. It's just the way the regulations were written, in an era when nobody was thinking about absorption. But we should think about it now. The labels are giving you a picture of the food that is technically true and functionally wrong. A second column, showing bioavailable nutrients adjusted for anti-nutrient content and standard preparation, would close the gap. Until then, here's the cheat code: **boil your spinach. Soak your grains. Sprout your beans. Ferment your cabbage. Drink vitamin C with iron-rich foods. Don't drink coffee or tea with meals when you're eating for minerals.** Or, alternatively, accept that you're getting less than the label promises and just eat a wide variety of foods and stop worrying about it. Either way, you should at least know. --- # Calories Should Be Divided by Ten URL: https://www.askgib.com/articles/calories-should-be-divided-by-ten Published: 2026-06-01 Reading time: 4 min > A calorie is so small that everything is measured in thousands of them. The unit is wrong. Divide it by ten and stop dealing with four-digit numbers for a sandwich. A banana has 105 calories. A bowl of pasta has 600 calories. A daily recommended intake for an adult is 2,000 calories. These are big numbers. They sound big because they are big. You will eat approximately **730,000 calories** this year. Over a lifetime, you will consume roughly **60 million calories**. The numbers are absurd for a measurement that's supposed to help you make daily decisions about what to eat. There's a simple fix. Divide everything by ten. Use a new unit. Call it whatever you want, let's say a "diet unit," or just "10 cal." | Item | Current calories | Diet units (calories ÷ 10) | |---|---|---| | Apple | 95 | 9.5 | | Banana | 105 | 10.5 | | Coffee with milk | 50 | 5 | | Sandwich | 400 | 40 | | Bowl of pasta | 600 | 60 | | Pizza slice | 285 | 28.5 | | Beer | 150 | 15 | | Daily intake (adult) | 2,000 | 200 | | Annual intake | 730,000 | 73,000 | *Same information. Smaller numbers. Easier to track mentally.* A banana has 10.5 units. A bowl of pasta has 60 units. Daily intake is 200 units. Your year is 73,000 units. These numbers are tractable. They live in the range your brain is actually good at. The calorie you see on every food label is technically a **kilocalorie**, 1,000 of the small calories that scientists use in physics and chemistry. The unit on food labels has been misnamed since it was standardized in the late 19th century, and we've never corrected the error. Here's the joke. The "calorie" on a nutrition label is already a factor-of-1,000 bigger than the actual unit it claims to be. We collectively decided that the real calorie was too small for everyday use, so we silently switched to a unit a thousand times larger and kept the original name. Then, having made that adjustment for usability, we proceeded to live in a world where everyday foods are measured in three-digit numbers and daily diets in four-digit numbers. It's like measuring distance in inches when you could use feet. We did half the conversion. We should finish it. The right scale for any common unit is one where typical daily amounts land in the 1-1,000 range, ideally in the 10-500 range. Currency: a coffee is $5, a meal is $30, a week of groceries is $150. Easy numbers. Distance: a walk to the store is 800 meters. Your commute is 12 kilometers. Easy. Mass: a bag of rice is 1 kg. Your weight is 70 kg. Easy. Calories: a snack is 200, a meal is 800, a day is 2,000. **Too many zeros.** Divide by 10 and: Snack: 20 units. Meal: 80 units. Day: 200 units. Workout burn: 30-50 units. Cookie: 5 units. Beer: 15 units. These numbers fit in your head. You can do the arithmetic without effort. "I had 80 for breakfast, I have 120 left for the day" is faster than "I had 800 for breakfast, I have 1,200 left for the day." Less cognitive load. Better decisions. More natural. The standard objection is that switching units is hard. Everyone is used to the current numbers. Every package, every recipe, every app, every restaurant menu is calibrated to the existing scale. Same objection that existed for switching from imperial to metric. The countries that switched are fine. The country that didn't (the US) is still using ounces, gallons, and Fahrenheit and complaining about it constantly. The switching cost is real. But it's a one-time cost, and the unit you end up with is better forever. Imagine if every nutrition label printed both numbers for ten years, then switched to the new unit alone. After a decade, nobody under 30 would even remember the old system. The transition would be uncomfortable. The destination would be better. | Unit reform that succeeded | Note | |---|---| | Imperial → Metric | Most of the world (except US) | | Fahrenheit → Celsius | Mostly Celsius globally now | | Calorie → kilocalorie (silent rename) | Already happened ~1890s | *The calorie was already re-scaled by a factor of 1,000 once. We can do it again.* The numbers people interact with daily would be smaller. Food labels would be easier to scan. Daily totals would be easier to track. The mental math of "should I eat this" gets simpler when the numbers involved are tens and hundreds instead of hundreds and thousands. For people tracking calories, and there are tens of millions of them, on apps like MyFitnessPal, Noom, Carb Manager, the new unit would make the whole exercise less psychologically punishing. Counting "20" for a snack feels different from counting "200" for the same snack. The new unit lets you feel like the snack is small. The current unit makes you feel like every meal is a major event. For policymakers, the new unit would make public health messaging clearer. "Eat 200 a day" is a cleaner target than "eat 2,000 a day." The proportions are the same. The friction is lower. The food industry doesn't want to relabel every product in the world. Regulators don't want to coordinate the change across borders. Consumers don't want to relearn the rules. The benefits accrue diffusely; the costs are concentrated on the industries that have to do the work. It's the same reason America still uses ounces and miles. The same reason we still count years from a religious event. The same reason your keyboard is QWERTY instead of Dvorak. The world is full of small bad standards. They persist because the alternative isn't worth the coordination cost of switching. The calorie is too small for daily use, and we already know it, because we silently replaced the actual calorie with a 1,000x-larger unit a century ago and kept the name. We should finish the job. Divide by ten. Live in a world where lunch is 70, not 700. Where a day is 200, not 2,000. Small change. Real improvement. Won't happen. But at least you can do the conversion in your head. --- # Time Is Broken URL: https://www.askgib.com/articles/time-is-broken Published: 2026-05-25 Reading time: 2 min > We're 2026 years into what, exactly? One religious event. The only honest unit is the day. So stop counting years and count days. You're not 27. You're day 9,866, and it moves every morning. We're 2026 years into... what, exactly? One religious event. That's the whole reason for the number. We measure the age of the universe with a calendar that started at a guy's birthday. Think about how dumb the units are. A *year*, one lap around the sun. A *month*, roughly the moon, except not really, so we faked it. A *week*, completely invented. Seven for no reason. Stacked on top of each other. None of them line up. That's why we need leap days to duct-tape the whole thing together. Here's the only honest unit. The **day.** One rotation of the Earth on its axis. That's it. That's real. You can feel it. Everything else is a story we agreed to. So here's the proposal. Stop counting years. Count **days.** Pick a real zero. The Big Bang. Or the start of human history. Doesn't matter, pick one, lock it, count forward. You're not "27 years old." You're **day 9,866.** Make it concrete. I was born June 9, 1999. In years, I'm 27. A number that means nothing and updates once a year. In days? I'm **9,866 days old.** And that number moves every single morning. It's alive. You watch yourself add up. Even better, I'm about to cross **10,000 days.** That's a real milestone. "27" is a shrug. "10,000 days" is a party. Now, the obvious objection. What about the huge numbers? Count from the Big Bang and you're sitting somewhere in the *trillions* of days. Nobody memorizes a trillion. So you do exactly what we already do with years. We don't say "the year two thousand and twenty-six." We say "twenty-six." We chop the front off and keep the part that moves. Same here. Lop off the cosmic prefix nobody needs. Keep the tail. The end of the number is the only part that actually changes in a human lifetime, so the end of the number is the only part you carry. Memorable. Personal. Yours. And the birthdays get *better.* 365 isn't sacred. It's one lap. Why throw a party for a lap? Pick real milestones instead. Every **1,000 days.** Every **500.** Every **100** if you're feeling generous. You'd celebrate way more often. Which, let's be honest, is a goldmine for everyone who sells cake, candles, dinners, and gifts. The entire birthday economy quietly triples overnight. So if you want this to actually happen? Don't pitch the philosophers. Pitch the party industry. Capitalism is going to *love* day-counting. It might be the one reform that actually passes. *(This is the part I want people to land on.)* The reason this matters isn't the math. It's the **mindset.** A year ties you to a season, a tax deadline, a birthday you didn't choose. A day ties you to nothing but the planet turning. Count days and you stop living on the calendar's schedule and start living on your own. Out at the poles there's barely a "day" at all. No clean sunrise, no lap that means anything. Time was never universal. We just pretended it was because it was convenient. We don't need the year. We needed it when farmers needed to know when to plant. We're not farmers anymore. You're not getting older. You're just adding days. Day 9,866. Still counting. --- # The Case for Slower Globalization URL: https://www.askgib.com/articles/the-case-for-slower-globalization Published: 2026-05-18 Reading time: 6 min > Borders are softening. Cultures are mixing. Mostly this is good. But there's a speed at which it starts erasing the places that made the journey worth taking in the first place. A few years ago, I walked into a small shop in Tokyo. I needed something I couldn't easily describe. I pulled out my phone, opened Google Translate, took a picture of the thing I was looking for, and held the result up to the shop owner. He was furious. He told me to leave. He didn't want foreigners in his shop, didn't want a phone shoved at him instead of an attempt at Japanese, didn't want any part of this interaction. In the moment, I was annoyed. I felt judged unfairly. I was trying. Later, it took me a while, I came around to seeing his point. I had walked into his shop as a foreigner and removed every human element of the encounter. No greeting in Japanese. No fumbling attempt at the language. No acknowledgment of him as a person whose space I was entering. Just a phone shoved at his face with a translation he didn't ask to participate in. He had every right to be annoyed. This article is about what we owe places we move through, and the speed at which globalization is happening, and why both matter more than most discussions of immigration ever admit. Borders softening has been, on balance, an enormous good. The ability to move where the opportunity is, to study abroad, to build careers across continents, to marry across cultures, to visit family thousands of kilometers away, this is one of the great achievements of the modern age. It has lifted hundreds of millions out of poverty. It has made cities more interesting, more dynamic, more cosmopolitan. Lebanese restaurants in Mexico City, Vietnamese pho in Berlin, Korean fried chicken in São Paulo, none of this existed a generation ago at the scale it does now. That interconnection has produced more wealth, art, knowledge, and human flourishing than any alternative we know how to build. Anyone who argues for completely closed borders or a return to ethnically homogeneous nation-states is arguing for a worse and poorer world. The case for openness is real and I am making it. What I want to argue is different. Openness has happened at a speed and scale, in some places, that has overrun the capacity of receiving communities to integrate the people arriving. That's a different problem from the principle of openness itself. And it's worth taking seriously without tipping into hostility toward immigrants as people. A neighborhood, a city, a country, is more than its physical infrastructure. It's a set of shared expectations about how people behave, how transactions work, what's polite, what's rude, what the small rituals of public life look like. These shared expectations are what make a place feel like that place. When people move into a community, they bring their own expectations. This is fine, it's how cultures evolve and enrich each other. Over a few generations, the receiving culture absorbs some of the new patterns, the arriving population adopts some of the local ones, and the new equilibrium is usually richer than either starting point. This is how it has worked, historically, in waves that arrived at a manageable pace into communities with the institutional capacity to integrate them. When the pace gets too fast, or the scale too large, or the receiving institutions too weak, integration doesn't happen. You end up with parallel communities living in the same physical space but operating on different norms. The local expectations get diluted, then forgotten. The arriving expectations don't displace them, they just coexist, creating friction. | Slow integration (~1% population / year) | Fast integration (~5%+ population / year) | |---|---| | Newcomers learn local norms | Parallel communities form | | Local language and customs persist | Local norms get diluted but not replaced | | Cultural mixing produces richer hybrid | Receiving institutions can't absorb | | Receiving institutions absorb arrivals | Both populations feel grievance | *It's not the principle that breaks. It's the pace.* This is not anyone's fault, individually. The immigrants didn't ask to be brought in at a pace that overwhelmed the integration capacity. The locals didn't ask for their neighborhood to change faster than they could adjust. It's a system-level failure, and every party has reason to be unhappy about it. The shop owner who threw me out wasn't reacting to me personally. By the time I walked in, he had been polite to thousands of people who couldn't be bothered to learn *konnichiwa*. I was just one more. Tokyo is one of the most modern, technologically advanced cities in the world, and it has resisted full globalization more deliberately than almost anywhere of comparable size. In 2026, it still feels distinctly Japanese. The shops are still mostly run by Japanese people. The norms of public behavior are still legible. The cleanliness, the quietness, the politeness, the orderliness, these aren't accidents. They are actively maintained by a society that decided, decades ago, that this is what they wanted their cities to be. Some of this maintenance is, frankly, unwelcoming. Restaurants that won't serve you if you're not Japanese. Apartments that won't rent to you. Shops where the owner kicks you out for using a phone. As an outsider, these experiences are uncomfortable. But would Tokyo still be Tokyo if it were as porous as London or New York? Would the quietness survive? The orderliness? The thousand small things that make Tokyo *Tokyo*? I don't think so. It would become a generic global city, indistinguishable from the parts of London and New York where local character has been worn away by sheer volume. It's not unreasonable for Tokyo to refuse that. If everywhere becomes available to everyone, on everyone's own terms, then nowhere is distinctively anywhere. The Tokyo I traveled to see only continues to exist because someone is still defending what it is. Open borders, fast: bad. Open borders, gradually, with integration capacity built in: good. International business, fast: builds wealth, builds trade. International business, so fast that local economies can't adapt and entire industries collapse without replacement: builds a cultural backlash, and rightfully so. Walk through the centre of any major city in Europe or America today. Now look at a photograph of the same street fifty years ago. It is barely the same place. Cities change, that's not a problem in itself. The problem is that the change happened faster than integration could keep up. People arrived in numbers the host society had no plan to teach. So they fell back on what they knew. And the world they had left, the one they were trying to escape, ended up being rebuilt in the place they came to. You are a product of your environment. A newcomer can't be judged for not knowing the local customs if no one ever taught them. They can't learn what isn't offered. The failure is not the newcomer. The failure is the host society that opened the door without putting anyone in the entrance hall. There's a speed limit on how much change a community can absorb without losing its coherence. It varies, small homogeneous countries have low limits; large heterogeneous cities have higher ones, but it exists everywhere. And the political consensus in most developed countries has been to either ignore the speed limit (the elite cosmopolitan view) or to deny the legitimacy of the change altogether (the populist nationalist view). Neither is right. The honest position is that openness is good, integration is necessary, and pace matters. Communities have the right to a coherent identity that persists across generations. Newcomers have the right to a fair chance to integrate. And the institutional infrastructure to make this work, language education, civic instruction, housing capacity, employment integration, has to be built before the flow arrives, not in panic afterward. > **Integration capacity** = housing + language services + civic education + employment integration + receiving culture's willingness > > When all variables are strong → high capacity, fast immigration sustainable. > When any variable is weak → capacity limited, immigration must slow to match. *Communities can absorb large numbers of newcomers only when the institutional infrastructure is built first. Most countries skipped the infrastructure step.* Immigration tied to receiving-community capacity. If a city's housing, schools, and integration services can absorb 10,000 new arrivals a year, the inflow is 10,000. If it can absorb 50,000, the inflow can be 50,000. The number is set by what works, not by political instinct. Mandatory integration support. Language instruction, civics, employment assistance, mental health resources, all front-loaded into the first two years. Free. Required. The state spends real money on making the arrival successful, because successful integration is what makes the whole system function. Cultural acknowledgment from arrivals. Not assimilation, nobody wants people to erase who they are. But acknowledgment that you are moving into a place with existing norms, and a baseline effort to learn them is part of the deal. The shop owner in Tokyo doesn't need me to be Japanese. He needs me to say good morning when I walk into his shop. That's a fair expectation. A real conversation about character maintenance. Communities are allowed to want to preserve their character. This is a legitimate civic interest, not a hateful one, and pretending it isn't only feeds the political backlash. The discussion should be about how to balance openness with character preservation, not whether character matters. The argument for globalization was always that the world would be better if everyone could move, trade, and learn freely. That argument was correct, and remains correct. The problem has been the speed. We built the infrastructure for moving people and goods across the world faster than we built the capacity to integrate the resulting flows. The local character of receiving communities has thinned, the arriving populations have not been adequately supported, and everyone has reason to be unhappy. The fix isn't closing borders. The fix is slowing down, building integration capacity, and being honest that places have a right to a coherent identity that survives the journey. I owe the Tokyo shop owner an apology I'll never deliver. He was right. We don't need less openness in the world. We need a version of openness that takes the human side of the receiving community seriously, instead of treating their resistance as bigotry to be overcome. Globalization should be a contract, not an invasion. And right now, in too many places, it's been operating as the second when it was supposed to be the first. --- # Too Many Doors, No Exit URL: https://www.askgib.com/articles/too-many-doors-no-exit Published: 2026-05-11 Reading time: 4 min > Canada didn't lose control of immigration by letting too many people in. It lost control by building a dozen ways in and no way out. Canada didn't lose control of immigration by letting too many people in. It lost control by building a dozen ways in and no way out. The word was always "temporary." Temporary workers. Temporary students. Temporary protection while a claim gets heard. That's normal. Every country does it. The problem is that nothing here is temporary anymore. People come through one door. When it closes behind them, they find another. Student to worker. Worker to asylum. Asylum to appeal to stay. The status keeps changing. The person never leaves. Look at the gap. Permanent immigration, the planned, points-tested, orderly kind, stayed flat for a decade. Somewhere between 185,000 and 483,000 a year, now cut back to around 390,000. That's the door with a lock on it. The other line is temporary residents. For most of the 2010s it sat under a million. Then it more than doubled in three years and peaked at 3.15 million in late 2024. It's only falling now because permits are finally expiring faster than Ottawa hands out new ones. That isn't a door anyone designed. It's a wall that fell down, and a government scrambling to rebuild it. A third of a million study permit holders in 2015. Over a million by 2023. Schools figured out that a foreign student pays three to five times domestic tuition, and the provinces had quietly stopped funding them. So the colleges went hunting. Mostly the public ones, despite what the politicians said. Twelve Ontario public colleges tripled their international numbers. An immigration lawyer called it exactly what it was. A cash cow. And the "school" part became optional. Statistics Canada found that nearly one in five study permit holders may not even be attending the institution on their permit. Go stand at Yonge and Eglinton at lunchtime. You'll see a crowd of students waiting on the sidewalk to sign an attendance sheet, so a piece of paper can say they showed up. Then they leave. No class. Just the signature. I've watched it happen for years. The fake-school part isn't a rumor either. In 2024 the government uncovered more than ten thousand fraudulent acceptance letters. One agency in Jalandhar alone filed roughly seven hundred fake applications over four years. The immigration minister's own word for the colleges churning out these degrees was "puppy mills." The tell is the date. May 15, 2024. That's when Ottawa quietly stripped the automatic work permit from the public-private "partnership" programs. You don't kill a perk like that unless you already know it was being farmed. Sixty-four thousand asylum claims in 2019. A hundred and seventy-three thousand in 2024. Look at where it climbs. Right as the temporary stack got too big to keep extending. People who came to work or study, running out of runway, did the rational thing. They claimed asylum. Because a claim buys years. Years of hearings, appeals, work permits, healthcare. The backlog is so deep that the wait itself is the reward. Then look at the last point. In 2025 claims finally dropped, by about a third, the moment Ottawa tightened visa screening on the travellers most likely to flip to a claim. Which tells you the whole thing was always manageable. They just had to decide to manage it. This isn't most of them. Most temporary residents work, pay rent, pay tax, and would go home if home were the only option. But the door-switching is real, and it's the part nobody designed and nobody could stop. Because there is no real exit. The minister got the obvious question in committee. Your own documents say 4.9 million permits and visas expire by the end of 2025. How will you know they left? His answer: the vast majority leave voluntarily, and that is what is expected. Expected. Not counted. Expected. Don't let the two numbers trip you up, because the government would love it if they did. Three million is how many temporary residents are actually here. The 4.9 million is how many permits and visas expire over about a year, which is a different thing, because it counts documents and not people, and it sweeps in tourist visas held by people who never lived here. But that mess is the point. There are so many overlapping permits, visas, and statuses that nobody, not even the department, can hand you one clean number. And a number nobody can pin down is a number nobody can be held to. Here is why "expected" was the only answer he had. Canada barely tracks who leaves. There is no exit gate, no scan on the way out the door. The border agency stitches together airline lists and a bit of data the Americans share at the land crossing, and then admits it cannot add it up to say who actually went home. They are only now, in 2026, building a pilot just to flag whether a temporary resident is still in the country. They are inventing the ability to answer the question after they already let the permits lapse. A government that could prove millions left would shout it from the rooftops. This one shrugs and takes people at their word. We do know the concrete pieces. Three quarters of a million study permits expired by the end of 2025, and better than half a million work permits on top of that. We know Canada removed fewer than 20,000 people that year. And we know the government's own estimate is that up to 500,000, maybe 600,000, are already here with no status at all. So when the temporary count drops, we don't actually know what we are looking at. Maybe they left. Maybe they slipped into the cash economy. The honest answer is the one nobody in charge will say out loud: we don't know. We built a country that counts you on the way in and loses you on the way out. It's the unfairness built into the design. Canada runs exactly one honest immigration system. Permanent residency. Points, skills, a queue, a wait, a yes or a no. And that is the door they're now shrinking. Meanwhile the backdoors stay open, because closing them is hard and nobody wants the fight. The diploma mills. The expired permits. The asylum backlog that rewards delay. So the person who did it right, applied, qualified, waited, gets squeezed. And the person who strung together three programs in a row gets to stay, because removing them is too much trouble. We are punishing the orderly to avoid the work of fixing the disorderly. The fix was never cruelty, and it was never a wall. It's the boring stuff. Fewer doors. Honest schools. A working exit. And the nerve to say that "temporary" has to mean something, or it means nothing. For years it meant nothing. We're only now finding out how hard it is to make it mean something again. --- # Give Homeless People Card Terminals URL: https://www.askgib.com/articles/give-homeless-people-card-terminals Published: 2026-05-04 Reading time: 4 min > Cash is dying. Charity is dying with it. The fix is municipal payment terminals, but really, in a country this rich, there shouldn't be a need for them at all. You walk past a homeless person on the street. Twenty years ago, you might have had a few coins in your pocket. You might have dropped one in the cup. The transaction was simple. They had something they didn't have before. You felt slightly less complicit in walking past suffering. Nobody had to verify anything. Now your wallet has no cash in it. Your transit pass is on your phone. Your coffee is paid by tapping your watch. Your groceries are scanned at a self-checkout. There are no coins, because there's no need for coins. The entire infrastructure of small generosity that used to operate between strangers, a quarter here, a dollar there, has quietly disappeared. The homeless person is still on the street. The cup is still there. But the transaction that used to fill the cup doesn't work anymore. This isn't an abstract problem. In the past decade, donations to street-level homeless individuals have collapsed in cashless societies. Sweden, the leader in digital payments, saw informal charity drop dramatically in the 2010s as cash use fell below 10% of transactions. The same trend is visible everywhere cash has retreated: South Korea, the Netherlands, increasingly Canada and the UK. Some homeless people have tried to adapt. A small number have set up GoFundMe accounts, posted Venmo handles on cardboard signs, accepted bank transfers. This works for a few who are tech-savvy and visible enough to get attention. It does not work at scale. Most people walking past will not stop to take out their phone, scan a stranger's QR code, sign into their banking app, and complete a transfer for one dollar. | Country | Cash share of transactions | Informal street giving | |---|---|---| | Sweden | <10% (2020s) | Collapsed alongside cash | | South Korea | <15% | Sharp decline | | Netherlands | <20% | Sharp decline | | Canada | 20–30% and falling | Steadily declining | *Where cash retreated, the small change that filled cups went with it. The technology that replaced cash never built a path for street-level giving.* What if every homeless person, in cooperation with their municipality, was given a small contactless payment terminal? The kind Square or Stripe ships for free to small businesses. City verifies recipient is in housing assistance system City issues authenticated terminal + linked account Passerby taps card or phone, 10¢ pre-set transferred Funds go to recipient's account, with city oversight *Same authentication standards as any merchant terminal on the Visa/Mastercard network. Tap and walk.* The terminal sits in their cup or on their sign. A passerby who wants to donate taps their card or phone to the terminal. A pre-set amount, say 10 cents, is transferred to a verified account. People used to drop a penny in the cup without thinking; this is the digital version of that. The whole transaction takes about a second. There's no QR code to scan, no app to open, no contact information exchanged. You tap and walk. The crucial mechanism is that the terminals are issued and authenticated by the municipality. Each terminal is registered to a specific individual through a city program. The account it transfers to is opened with help from the city's social services. This solves the two biggest objections people raise. First, security. Without municipal authentication, the obvious worry is hacking, a malicious terminal that captures card details. With city-issued, regularly audited terminals on the standard Visa/Mastercard contactless network, this risk is no higher than tapping your card at any other small business. The infrastructure exists. It's the same as any street vendor or food truck. Second, fraud. People worry about scams, someone with a terminal who isn't actually homeless, or someone using the money for things the donor wouldn't want to fund. The municipal authentication addresses the first; the terminal is only issued through a process that confirms the individual is in the city's housing assistance system. The second concern is real but is also true of cash; a dollar given is a dollar that recipient spends however they want, and most reasonable models of charity accept that. The terminal model has been tried, with various levels of success, in Amsterdam, Stockholm, parts of London, and a few US cities. The technology works. The deployment has been small and underfunded. It deserves to be the default. In a country with the per-capita wealth of any developed economy in 2026, there should not be homeless people in significant numbers. The fact that there are, in every major Western city, is not a result of insufficient resources. It is a result of how those resources have been allocated. Most developed countries spend somewhere between 30% and 50% of their GDP on government. A large fraction of that is on social programs. Housing, in particular, is the most obvious lever, building it, subsidizing it, ensuring it's available to people who can't access market housing. Yet in San Francisco, in Vancouver, in Sydney, in Paris, in Berlin, in Toronto, homelessness has been growing for two decades. Not because resources are scarce. Because the political will to build housing at the scale needed has been absent, and because the systems that should catch people before they end up on the street, mental health care, addiction treatment, transitional housing, employment assistance, are underfunded, fragmented, and often hostile to the people they're supposed to serve. A homeless person with a payment terminal is a better outcome than a homeless person with an empty cup. But the existence of the homeless person is the actual scandal. The terminal is a small humane patch on a much larger civilizational failure. The terminal program should be implemented as a default municipal service in any city with a homeless population. The technology costs are trivial. The administrative overhead is small. The benefit to people living on the street is meaningful. Cities that haven't done this should. But the actual fix is upstream. Build housing, public, social, transitional. Fund mental health care that people can actually access, not just have a phone number for. Treat addiction as a health condition and not a criminal one. Make the transition from temporary crisis to permanent stability easier, not harder. None of this is mysterious. Finland did most of it and reduced rough sleeping by over 80% with a "Housing First" policy that has been studied, copied, and successfully implemented elsewhere. There is no developed country in 2026 where the existence of widespread street homelessness is a financial problem. It's a political problem. It's a choice. And the terminal is a humane response to a situation that shouldn't exist in the first place. We've spent twenty years building the most efficient payment infrastructure in human history. Trillions of dollars move through it every day, at near-zero cost, faster than the previous generation could have imagined. We have not figured out how to include in this system the most visible, most vulnerable members of our communities. They've been left behind by the same modernization the rest of us have benefited from. The cup is empty because the system that used to fill it doesn't work in this economy. Give them terminals. It's a small thing, and it costs almost nothing. And then build the housing. Fund the care. Solve the actual problem. The terminal is a patch. The empty cup is a scandal. We can do both. --- # The People's Bank URL: https://www.askgib.com/articles/the-peoples-bank Published: 2026-04-27 Reading time: 6 min > A bank that doesn't lend, doesn't speculate, doesn't have shareholders. Just a place to store money. With modern infrastructure, this is basically free to run. You deposit your paycheck at a bank. The bank takes your money, lends it to a stranger at 6%, pays you 0.5%, and keeps the 5.5% spread. Then they charge you a monthly account fee, an overdraft fee, a wire transfer fee, an out-of-network ATM fee, a "maintenance" fee, and a "paper statement" fee. Then they spend a portion of their profits on a Super Bowl ad telling you how much they care about your financial wellbeing. You provide the capital that funds their entire business. They charge you for the privilege. There should be another option. **Business one: storage.** You give them your money. They hold it. You can take it out, transfer it, spend it via a card. This is the part you actually use. Storing digital balances is, in 2026, one of the cheapest things any business can do. The marginal cost of one more account is approximately zero. **Business two: lending and investment.** Your deposits are aggregated and lent out, and used as collateral for the bank's own investment activity. Mortgages. Business loans. Bond positions. Derivatives. Sometimes outright speculation. This is where the bank makes its money, and where the risk lives. When a bank fails, Silicon Valley Bank in 2023, Credit Suisse the same year, dozens of regionals in 2008, it never fails because the storage business broke. It fails because the lending made bad bets, and the depositors' money was the collateral. | What happens with your $10,000 deposit at a commercial bank | | |---|---| | Your deposit | $10,000 | | Bank keeps in reserve (10%) | $1,000 | | Bank lends out (90%) at 6% | $9,000 | | Borrower repays | $9,540 over the year | | Bank pays you (0.5%) | $50 | | **Bank keeps the spread** | **$490 (5.5%)** | *You funded a $9,000 loan. The bank earned 10x what they paid you for that funding.* Why are these bolted together? Historical accident. In the 18th and 19th centuries, banks needed deposits to fund their lending, and depositors got safekeeping in exchange for the implicit risk. The trade-off was reasonable then. Now, with central bank deposit insurance, modern payments infrastructure, and digital ledger systems, the bolt could be unbolted. A nonprofit institution. Owned mutually by its account holders, the way a credit union is. No outside shareholders. No bonus pool for executives. **Account types: one.** A regular deposit account. You put money in, take money out, spend via debit card or digital wallet, wire to other accounts. That's the entire product. **Interest paid:** the central bank's overnight rate, minus a small operating cost (maybe 0.10%). When the Fed pays 5.25%, depositors get roughly 5.15%. When the Fed pays 0%, depositors get 0%. The pass-through is honest. **Lending:** none. The full deposit base sits at the central bank or in short-term government securities. The principal is as safe as the government is solvent. **Fees:** essentially none. No monthly maintenance. No overdraft (the account simply rejects transactions that would overdraw it). No wire fees between People's Bank accounts; outbound wires charge the actual interbank cost, which is cents. No ATM fees on the own network; reciprocal arrangements elsewhere. **Infrastructure:** modern, digital-first, almost entirely automated. A debit card. A mobile app. A web interface. Customer service for problems. Branches are optional and few. | | Commercial bank | People's Bank | |---|---|---| | Ownership | Shareholders | Account holders | | Lending | Yes (extensive) | None | | Interest rate to depositors | 0.5% (typical) | Central bank rate minus 0.10% | | Fees | Many | Near zero | | Risk | Real | None (deposits in central bank only) | | Profit motive | Maximize | None | | Branches | Hundreds | Few or zero | | Headcount | Thousands | Small operations team | That's the entire institution. The trading desks, risk management, loan underwriting, marketing budget, executive suite, wealth management arm, all of that goes away. What's left is the part you actually use. It almost does. Credit unions are close. So are postal banks in some countries, and state-owned savings institutions in parts of Europe and Asia. But each has been compromised. Most US credit unions lend extensively, they're not really storage-only. Postal banks in Japan (Yucho), Germany (formerly Postbank), and elsewhere have varied wildly in quality. State-owned banks frequently get captured by politics or pushed into lending to favored sectors. | Closest existing alternatives | Note | |---|---| | Credit unions (US, UK, Canada) | Closest to the model, but still lend | | Postal banks (Japan Yucho; formerly German Postbank) | Varied implementation | | Wise | International transfers only, private | | Estonia digital banking | Closest to fully digital model | The pure version, mutually owned, storage-only, near-zero-fee, running on modern digital infrastructure, doesn't exist at scale. The closest things are fintechs like Wise (for international transfers) or Apple Cash and Cash App for small-balance storage. But these are private companies with shareholders who eventually want returns. The structure isn't mutual. The economics will force them to either become real banks or get acquired by them. What's missing is the institutional model: a real bank, large enough to be useful, structured so no outside party can extract value from it. **"But how would the economy fund itself?"** The standard banker's response. If everyone moved to a non-lending bank, where would loans for houses and businesses come from? From the lending banks, which would still exist. The People's Bank is an *option*, not a replacement. Anyone who wants the riskier deposit-lending system can keep their money at Chase or Citi. In practice, most people would split: daily expenses and emergency savings at the People's Bank, investment money at brokers, some lending exposure at a regular bank if they want it. The lending system would shrink, not disappear. And it would have to attract depositors honestly, with interest rates that actually compensate for the risk. Right now, banks pay 0.5% on accounts that fund 6% mortgages because most depositors don't understand they're financing the bank's business. With a clear alternative, that information asymmetry closes. **"Won't it fail when something goes wrong?"** No, because it has no investment portfolio that can go wrong. The deposits sit at the central bank or in short-term sovereign debt. By design, the most boring financial institution possible. No trading risk, no credit risk, no concentration risk. The only way it fails is if the central bank fails, in which case all bets are off across the entire economy. **"Who would run it?"** A small professional staff. Day-to-day operations of a storage-only bank in 2026 are mostly automated. The team would be smaller than most credit unions. The board would be elected by account holders, like credit union boards. Compensation: normal nonprofit salaries, not Wall Street salaries, because there's no Wall Street work being done. The most important effect is psychological. Right now, most people interact with the financial system as customers of a hostile counterparty. Every fee, every rate decision, every product is engineered to maximize bank revenue. You are not the customer. You are the raw material. A People's Bank flips this. You're a member-owner of the institution holding your money. The rate you receive is a pass-through of what the central bank pays. The fees you pay are operating costs, not profit. The institution exists to serve you, because there is no one else to serve. This is what credit unions were originally supposed to be. They started in the 19th century as cooperatives for workers and farmers who couldn't get fair treatment from banks. Most drifted toward bank-like behavior because they competed with banks on lending. The People's Bank would be deliberately constrained, no lending, no drift. For the financial system as a whole, a clean storage alternative would force commercial banks to compete more honestly. They couldn't get away with 0.5% deposit rates when an alternative pays 5.15%. They'd have to raise rates or shrink. Either outcome is healthier than the current model. This is technically possible. It is also politically nearly impossible to launch in most countries, because the banking lobby would fight it ferociously. Banks own a significant portion of national legislatures, especially in the US and UK. They will not voluntarily allow a competitor that disrupts their economic model. They'll argue it would destabilize the financial system, undermine credit availability, "starve the economy of capital," and on and on. Some arguments are made in good faith. Most are made by people whose jobs depend on the current model continuing. The cleanest path is probably not national legislation, but launching it as a credit-union-plus structure in a friendly jurisdiction with strong consumer protection law, modern digital infrastructure, and an open regulator. Estonia, Switzerland, possibly New Zealand or Singapore. Make it work small, prove the model, scale from there. Once the existence of a working People's Bank in one country is visible, the demand to replicate it elsewhere becomes hard to suppress. That's how Wise scaled the international money transfer market against bank resistance. It's how Vanguard scaled the low-cost index fund against the actively managed fund industry. The pattern is: build a working alternative in a friendly jurisdiction, demonstrate the user experience, let the political pressure for replication build. Banking is not a difficult product. The technology is straightforward. The infrastructure is widely available. The only reason most people are paying spreads and fees for basic money storage is that the banking industry has spent two centuries making sure no alternative exists. The People's Bank is what banking should look like once we strip out the parts that exist to extract value from depositors. It's mutual. It's transparent. It's boring. It pays you what your money is actually earning at the central bank. It doesn't sell your data, it doesn't speculate with your deposits, and it doesn't fund a marketing department to convince you that the way things are is the way things have to be. You shouldn't need to learn finance to avoid being financially exploited. You should just need an account. --- # You Shouldn't Vote on Your Own Paycheck URL: https://www.askgib.com/articles/you-shouldnt-vote-on-your-own-paycheck Published: 2026-04-20 Reading time: 2 min > If your income comes from the government, you shouldn't vote on how much it pays you. The most basic rule in finance and law, applied to the ballot box. Simple rule. If your income comes from the government, you don't vote on how much the government pays you. That is not cruelty, it is conflict of interest, the most basic rule in finance, law, and ethics. You can't sit on both sides of the table. A judge recuses himself. A board member with a stake leaves the room. The second it is votes, we pretend the rule vanishes. It doesn't. And here is the part nobody says out loud. This applies to everyone. Farmers voting on farm subsidies. Defense contractors voting on defense budgets. Civil servants voting on public pay. Pensioners voting to raise pensions. Welfare recipients voting on welfare. Same principle, no exceptions. The moment your ballot is also your invoice, it stops being a vote. It is a withdrawal. So, stated cleanly. People whose net lifetime contribution to the system is negative don't vote on their own benefits. Yes, that is a lot of people, on a lot of things. Not "poor people can't vote," not income today, but net contribution over a lifetime. The nurse who paid in for 40 years and now draws a pension paid her dues. The 28-year-old who has never net-contributed and votes himself a bigger cheque is the conflict, and it is the same conflict whether he is a welfare claimant or a subsidised farmer. And you can earn your way in. Put in the equivalent of half a full minimum-wage working life in tax, in Ontario today about CA$900,000 worth of contribution, and you are in for good, voting on everything. Don't want to work for it? Pay the lump sum and buy your seat. The franchise isn't taken away, it is earned, the way everything serious is earned. John Stuart Mill This isn't new. John Stuart Mill argued exactly this in 1861. Most democracies started with contribution-based franchises and only later went universal. We treat universal suffrage like physics. It isn't. It is a choice we made, recently, and we are allowed to ask if it is working. It also quietly settles the immigration fight nobody wins. The question was never where you came from, it is whether you have paid into the pot you are voting to split. Because right now the incentive runs one way. The people who benefit from a bigger state keep voting for a bigger state, and the bill lands on people whose vote doesn't get louder for paying more. Parties worked this out decades ago. You don't win the argument, you change the electorate. A vote should be a voice, not a paycheck. The day we let those two things merge, we stopped running a democracy and started running an auction. --- # Flying Needs Rules URL: https://www.askgib.com/articles/flying-needs-rules Published: 2026-04-13 Reading time: 5 min > Same seat, same flight, same day, two people paid prices that differ by 600%. Here are the rules flying needs. You book a flight last minute. It's $1,200. You sit down. Five seats around you are empty. You just paid for six tickets so the airline could *not* sell five of them. That's not a market. That's a hostage situation. The subway has a fare. You pay it. You ride. A train ticket includes your seat and your bag. Same with a long-distance bus. Same with a ferry. Same with the Eurostar. Flying is the only mode of transport where the price of the same seat, on the same day, can swing by 600% depending on when you click "book." Where your bag costs extra. Where your seat costs extra. Where the meal costs extra. Where the airline can sell more seats than the plane has and then bump you off the flight they sold you. None of this is illegal. That's the problem. There are no rules. I built a bottom-up cost model for 206 routes across every G20 country. Every cost component calculated from scratch. Fuel, aircraft lease, crew, maintenance, ground handling, overhead. Add a 15% profit margin. That's the fare. Here's what flying should cost, one-way, all-inclusive (bag, meal, seat, all fees): | Route | Distance | Should cost | Actually costs | |---|---|---|---| | São Paulo → Rio | 356 km | **$45** | $120 | | Seoul → Jeju | 451 km | **$51** | $110 | | Sydney → Melbourne | 706 km | **$58** | $195 | | Delhi → Mumbai | 1,137 km | **$68** | $135 | | New York → LA | 3,974 km | **$134** | $320 | | **London → New York** | **5,541 km** | **$290** | **$700** | | Sydney → LA | 12,060 km | $619 | $980 |
133 of the 206 routes can sustain economy fares **under $100 one-way**, fully all-inclusive, at a healthy 15% profit margin. The average across all 206 routes is **$131 one-way**. The average overcharge per ticket compared to what you actually pay today: **$139**. Multiplied across global passenger volumes, people are overpaying about **$1.2 trillion a year** to fly. When you pay $700 for a ticket, only a fraction of that pays for actually flying you somewhere. The rest goes to revenue management departments, which are entire teams of analysts whose only job is to figure out the highest price you'll pay before you give up and book. To GDS distribution fees, which are middleware companies like Sabre and Amadeus that charge $5 to $10 every time a ticket is sold through a travel agent or comparison site. To marketing budgets that spend billions convincing you their seat is somehow different from the other airline's seat. And to loyalty program accounting, which is bigger than you think. Kill those three line items and the price drops by 30 to 40%. No magic. Just stop paying for things that don't move planes. In 2020, American Airlines used its AAdvantage program as collateral for a $7.5 billion loan. The program itself was valued at $19.5 billion. More than the entire airline. Delta's SkyMiles? Valued at $26 billion. Bigger than Delta itself. American Airlines makes more pre-tax profit selling miles to Citi than from operating flights. Delta gets $7 billion a year from American Express for the right to issue SkyMiles to cardholders. The flying is the marketing. The miles are the business. And the miles themselves are a Ponzi scheme. Not in the legal sense, nobody's going to jail. But structurally. Banks buy miles from airlines at about 1.5 cents each. The airline books the cash as revenue immediately. The miles become a "liability" on the airline's balance sheet, but the airline controls how much they're worth. Over time, the airline quietly raises the redemption cost. What used to be 25,000 miles becomes 75,000 miles. Same flight. In 2023, American Airlines just deleted their award chart entirely. Now there's no fixed value at all. Your miles are worth whatever they say they're worth, on whatever day you try to use them. Who wins? People who cash out early. Who loses? The "loyal" customer who accumulates miles for ten years and finds out his free trip to Hawaii now costs three times as many miles as when he started. The credit card piece is even darker. The average household carrying a $5,000 balance at 24% APR pays about $1,200 a year in interest to earn miles worth maybe $150 in actual redemption value. The miles are the bait. The interest is the business. Great circle distance: 13,700 km. Actual route via Dubai or Amsterdam: 18,000 km. That's 4,000 extra kilometers of fuel, emissions, and your life, every single trip, because the airline makes more money funneling you through their hub.
Hub-and-spoke exists because airlines profit from it. Passengers hate it. The planet hates it. Air traffic control hates it. Under a fixed-fare model, there's no reason to force connections. Routes with demand get direct flights. Eliminate the unnecessary connections and global aviation emissions drop 15 to 25%, without changing a single piece of technology. Just by flying in straight lines. **Rule 1. Standardize the ticket.** A ticket includes your seat, one checked bag up to 23kg, one carry-on, a meal, and all airport fees. No "basic economy." No baggage upcharge. No seat selection fee. The price you see is the price you pay. The Eurostar already does this. Every subway on Earth already does this. It's not complicated. **Rule 2. Cost-reflective fixed fares.** A standard formula, published publicly, updated annually. Same route, same fare. Doesn't matter when you book or what device you search from. Last-minute travel costs the same as planning six months out. Carriers compete on reliability, cabin quality, on-time performance, route frequency. Not on who can extract the most money from a desperate traveler. **Rule 3. A Global Air Access Fund for thin routes.** Some routes (small islands, remote regions) won't break even at standard fares. Fund those through a 1.5% levy on global airline revenue, about $14.5 billion a year. Carriers bid to operate the routes at the standard fare. Same way rural mail delivery works in every developed country. Same way rural broadband subsidies work. This isn't new. This is just applying a model that already exists. The legal framework is already there. The Chicago Convention of 1944, the treaty that governs international aviation, already has the authority to standardize fares. It just hasn't been used. This isn't socialism. It's how every other essential transport network on Earth already operates. You need to fly home because someone in your family is sick. The ticket that was $300 a week ago is now $1,400. Same seat. Same flight. Same fuel cost. You're not paying for a service. You're paying for *being in a hurry*. The airline knows you have to fly, and they price-discriminate on your desperation. That's not a market function. That's extraction. And then, and this is the part that really gets me, you get on the plane and there are empty seats all around you. The flight is half full. The airline would rather fly a plane with empty seats than sell those seats at a price normal people can afford. That's not capitalism failing. That's capitalism working exactly as designed, to maximize revenue per passenger, not passengers carried. Subways don't do this. Trains don't do this. Buses don't do this. Only airlines. The world is more connected than ever, and the price of staying connected keeps going up. That's backwards. Half the planet has a life that requires a plane to maintain. Refugees, migrants, expats, students, remote workers, families spread across three continents. You can't tell people to move where the opportunity is and then price them out of ever going back to see their family. You can't accept immigration and then trap people once they arrive. It can't be a one-way street. If we're going to keep telling people the world is open, we have to make it actually reachable. $290 to fly New York to London. That's the real price. Everything above that is the airline charging you for the fact that you didn't know. Flying doesn't need to be free. It just needs rules. --- # The Subsidy Scams Nobody Talks About URL: https://www.askgib.com/articles/the-subsidy-scams Published: 2026-04-06 Reading time: 6 min > Diagnostic immobilier. Mandatory home security inspections. Permit consultants. Entire professions exist because the government created them, not because the market needs them. You sell your apartment in France. Before you can close the sale, you need a stack of paperwork. A DPE (energy performance diagnostic). A diagnostic amiante (asbestos). A diagnostic plomb (lead). A diagnostic gaz. A diagnostic électricité. An ERP (state of natural and industrial risks). For older properties, a CREP. For apartments in copropriété, a Carrez. If your DPE is class F or G, an audit énergétique. The bill, depending on the property: **€400 to €700 for an apartment, €1,100+ for an older house**. None of these inspections will actually fix anything. They produce documents. The documents get attached to the sale contract. Then everyone goes home. This is one of the largest hidden taxes on property in France. The diagnostiqueur immobilier is a job that was created, not discovered. There was no market demand. There was a regulatory mandate, and a profession appeared to extract fees against it. *Every developed country has dozens of these. The pattern is identical. Each step is hard to reverse once it happens.* Step one: A government identifies a real problem. Asbestos in old buildings. Lead paint. Energy efficiency. Workplace safety. Food handling. Fire risk. Step two: Instead of solving it directly, funding remediation, building public capacity, training enforcement officers, the government passes a law requiring private compliance. The owner must obtain a certificate. The restaurant must display a permit. The employer must complete a training. The contractor must hold a license. Step three: A new profession emerges around the certificate. People train, get licensed, and charge for the document. The fee is the product. The actual underlying problem, the asbestos, the lead, the safety risk, is incidental. The profession profits from the requirement, not from the solution. Step four: The profession lobbies to make the requirement more complex. More categories. More frequency. Higher minimum prices. Stricter penalties for non-compliance. They're a focused interest group; their opponents are diffuse property owners who don't even know the law exists until they need it. Step five: The requirement becomes culturally entrenched. The profession becomes "essential." Suggesting it's a tax in disguise gets you accused of wanting to harm safety, or the environment, or workers. The DPE, the most famous of the French diagnostics, costs **€90 to €160** for a typical apartment. Roughly 4 million are issued every year. That's a market of **€400 to €640 million annually**, just for one diagnostic. The total annual market for all French diagnostics immobiliers is several billion euros. *For a €300,000 apartment, you'll pay €1,500+ in compliance paperwork. The DPE alone has produced no measurable change in French residential energy use since it became mandatory.* According to housing ministry estimates published in 2025, **roughly 70,000 DPE certificates per year are fraudulent**. That's about 1.75% of all DPEs. Some diagnosticians manipulate the rating to help owners avoid new laws banning rental of class-G properties. Others just collect the fee without doing the work properly. The system has near-zero enforcement, because the regulator (the cofrac certification body) is staffed by people from the same industry. The result is a mandatory product, sold at scale, with significant fraud, that produces a certificate which gets ignored by most buyers. Most people look at the DPE rating, shrug, and proceed with the sale. The information rarely changes the price or the decision. It's a tax the buyer pays to a parallel industry of certificate-issuers. And the kicker: the DPE was introduced to drive energy efficiency improvements. French residential energy use per square meter has barely moved since it became mandatory. What has occurred is a €400-million-plus annual industry, employing tens of thousands of people, producing documents nobody reads. In Canada, **mandatory home inspections** before sale generate a $300 to $700 fee per transaction. The inspectors are licensed. The inspections are real, technically. The reports are rarely actionable. In Italy, **certificazione energetica (APE)** does the same thing as the French DPE, with similar issues. In Germany, the **Energieausweis** is the local version. In the UK, **HIPs (Home Information Packs)** were introduced in 2007, then abolished in 2010 after sustained criticism that they added cost without value. Across the world, **food handler permits**, **workplace safety certifications**, **business operating licenses for trivial activities** (yes, you need a license to be a florist in Louisiana), and dozens of similar requirements have created small industries of certificate-issuers. None of them invented themselves. They exist because a law required them. Not all of them are useless. Proper electrical inspections do catch fires, proper food handling certifications do prevent food poisoning. But many are pure rent extraction. Once a profession exists around a requirement, it becomes politically impossible to dismantle the requirement. The workers who depend on it lobby hard to keep it. Their existence becomes the justification for the rule. In the past five years, regulators across the EU and the US have introduced cybersecurity compliance requirements: GDPR for data protection, the NIS2 directive for "essential service" companies in the EU, HIPAA in the US, PCI-DSS for payment processing, ISO 27001 for everything else. Each requires certification, audit, and ongoing compliance work. An entire profession of "compliance consultants" emerged within a decade. They charge five-figure fees to small companies for documents that are often lightly customized templates. The EU compliance industry alone is worth tens of billions of euros a year. Despite all this compliance work, **the rate of significant data breaches has continued to rise** every year. The regulations did not stop ransomware. They did not stop major breaches at Equifax, Marriott, Capital One, Optus, Latitude, Solarwinds, or any of the dozens of others. They created a parallel economy of consultants and auditors who get paid to confirm that companies have the right paperwork on file before they get hacked anyway. The cyber-security insurance racket is the next layer. The insurance has exclusions for "state-sponsored attacks," which insurers unilaterally define to include essentially any sophisticated ransomware group. When a real breach happens, the insurer denies the claim. The company loses anyway. The premium was already collected and invested. We covered this in the insurance piece, but here it is again, attached to a fresh mandatory product. | Compliance industry | Annual revenue | Underlying outcome | |---|---|---| | French DPE | €400M+ | ~0% measurable change in residential energy use | | EU GDPR / NIS2 compliance | €30B+ | Breach rates continuing to rise | *The compliance industry grows. The underlying problems don't shrink.* Same pattern. Mandate, profession, fee extraction, no impact on the underlying problem. In France, the total cost of compliance, certification, and mandatory paperwork on a typical apartment sale runs **roughly €1,000 to €2,000** when you add it all up, diagnostics, notary fees, registration fees, and the small army of intermediaries each transaction requires. For a €300,000 apartment, that's about half a percent of the sale price, going to people whose role is paperwork. Scale that across the entire economy, every sale, every lease, every business setup, every employment contract, every permit, every license, and you're talking about hundreds of billions of euros a year in advanced economies. Money that doesn't produce anything. It produces paper, that satisfies a regulator, that justifies a profession, that lobbies to keep the requirement in place. This is the hidden tax nobody talks about. It doesn't show up on your income tax statement. It shows up in the cost of doing everything. Every transaction is slightly more expensive. Every project takes slightly longer to start. Every product costs slightly more because the producer paid the parallel economy of certifiers along the way. The people who do these jobs are mostly decent. They followed a career path the government created. They didn't write the rules. But collectively the system is a transfer from productive activity to paperwork production, and most countries have a *lot* of it. The honest answer: very little can be done about these professions once they exist. The constituency to dismantle them is weak. The constituency to defend them is well-funded and focused. Property owners individually have small stakes; the diagnostician profession collectively has large stakes. The math favors the profession. First, build a regulatory hygiene practice. Every five years, every mandatory certification should be reviewed against the outcome it was supposed to produce. If the DPE hasn't reduced energy use, it should be reformed or abolished. If the home inspector industry hasn't improved housing safety, the requirement should be revisited. This requires political will that almost nowhere has. But it's the only honest fix. Second, when new requirements are introduced, **the law should specify a sunset clause**. The requirement exists for ten years, then expires unless reauthorized by a vote requiring the regulator to demonstrate measurable outcomes. Some good legislation already works this way. It needs to be the default. Without these mechanisms, every developed country accumulates parallel professions, like coral reefs, decade after decade. Each individually small. Collectively enormous. Untouchable as a group. You don't see these jobs on the news. They don't make headlines. There's no scandal. There's just a slowly accumulating cost on every transaction, every property, every business, every life, collected by professions that exist because the law requires them to exist. The diagnostic immobilier. The home inspector. The compliance auditor. The permit consultant. The food handler trainer. The mandatory drug counselor. The state-required notary stamp on a transaction that didn't need a notary fifty years ago. Each individually defensible. Collectively absurd. This is what people mean when they say the modern economy is bloated. They don't mean wages are too high. They mean we've built a layer of paperwork on top of every real activity, employed millions of people to produce that paperwork, and the paperwork mostly doesn't accomplish what it was supposed to accomplish. You're paying for it. You're paying for it on every transaction, every year, for your entire life. And there's no line item that tells you so. --- # The End of Generalist Doctors URL: https://www.askgib.com/articles/the-end-of-generalist-doctors Published: 2026-03-30 Reading time: 6 min > AI diagnoses skin cancer better than 80% of general practitioners. It reads x-rays better than most radiologists. The GP referral system is about to be replaced by a phone. You wake up with a rash on your arm. You book a GP appointment. The wait is two weeks. You go in. The GP looks at the rash for thirty seconds. They say it's probably nothing, but they want to refer you to a dermatologist just to be safe. The dermatologist's wait is six weeks. You spend two months mildly worried about a rash that turns out to be eczema. Now run the same scenario in 2026. You open an app. You take a photo. The AI tells you, with reasonable confidence, that it's eczema. It tells you when to worry. It tells you what to buy at the pharmacy. The whole thing takes ninety seconds and costs nothing. This is the end of the general practitioner as we know them. And it's already happening. A 2025 meta-analysis published in BMC Primary Care looked at AI diagnostic accuracy for skin conditions compared to general practitioners and dermatologists. GPs diagnose skin lesions correctly in the **20 to 40% range**. Dermatologists hit **around 85%**. AI systems, in a separate 2025 meta-analysis in Medicina pooling over 70,000 test images, reach **91% sensitivity and 88% AUROC** for melanoma detection. | Diagnostician | Skin lesion accuracy | |---|---| | General practitioner | 20–40% | | Dermatologist | ~85% | | **AI (pooled meta-analysis)** | **91% sensitivity / 88% AUROC** | *Source: BMC Primary Care 2025 and Medicina 2025 meta-analyses; Cureus 2025 study. That's specialist-level performance on a phone.* A separate 2025 study in Cureus compared ChatGPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro against board-certified dermatologists on 30 clinical cases. The conclusion: AI models showed diagnostic accuracy **comparable to and sometimes exceeding** the specialists, particularly on rare and complex cases. Radiology is even further along. A 2025 review in the journal Medicine reports AI improving accuracy and **reducing diagnostic time by approximately 90%** in radiology and pathology under research conditions. The technology is already in clinical use across Europe for detecting pulmonary nodules, intracranial bleeds, fractures, and early-stage cancers. The FDA has approved over 800 AI-enabled medical devices, most of them in radiology. And the gap is widening. Every year, the AI gets better. The general practitioner does not. They take vitals. They listen to your symptoms. They look in your ears, nose, throat. They feel for lumps. They run through a mental decision tree: this symptom plus that symptom equals one of these five possibilities. Then they either prescribe (90% of cases) or refer (10% of cases). The whole appointment averages 7 to 15 minutes in most developed countries. This is pattern matching. The GP is a human-shaped interface to a decision tree they memorized in medical school and have been refining for twenty years. The decision tree is now in software. The pattern matching is now in software. The prescribing is increasingly in software (electronic prescribing platforms are already standard). The referral is increasingly in software (AI triage systems that route patients directly to the right specialist are being deployed across the UK, Israel, Singapore). What's left for the GP to do that the software can't do? Three things. Physical examination, hands on the body, ears in the ear, listening to a chest. Building trust with the patient over time. And signing off on prescriptions that the legal system still requires a human signature for. These are real, but they don't justify a fifteen-minute appointment after a two-week wait, and they don't justify the cost structure of the entire primary care system. The instinct, when reading this, is to recoil. We don't want robots replacing doctors. We don't want our health in the hands of an algorithm. We want a human who looks us in the eye and tells us what's wrong. | Country | Average wait for specialist care | |---|---| | Canada | **27.7 weeks** (2023) | | UK (NHS routine) | 18+ weeks | | France | Variable, often months (médicaux deserts) | | Australia | 4–12 weeks | | US (with insurance) | 2–4 weeks | *Source: Fraser Institute (Canada), NHS, various national statistics. The romantic vision of the family doctor is dead. What's left is a system most people can barely access.* In Canada, the median wait from a GP referral to actual treatment is **27.7 weeks** as of 2023, the longest in three decades. In the UK, NHS waiting lists for routine treatment hit a record 7.6 million people in 2023. In France, the "déserts médicaux", medical deserts with no available GP, now cover **a third of the country** by population. In rural areas across the developed world, GPs are retiring faster than they're being replaced. The population is aging, demand is rising, and the supply of doctors is fixed by training capacity that takes a decade to expand. The current system is not delivering attentive, personalized human care. It's delivering rushed, overworked, often impossible-to-access human care. The romanticized version of the GP, the family doctor who knew you and your family for thirty years and made house calls, has been dead for a long time. What's left is a person in a white coat who has fifteen minutes to look at you, refers you to someone else, and bills your insurance. Against that reality, an AI that gives you an accurate diagnosis in ninety seconds, available 24/7, at near-zero cost, is not a downgrade. It's a massive upgrade. The argument isn't AI versus the ideal doctor. It's AI versus the doctor you can actually get an appointment with. The **specialist** survives. Cardiothoracic surgeons. Interventional radiologists. Pediatric oncologists. Anyone whose work requires a deep narrow expertise plus hands-on physical intervention. AI assists these people, doesn't replace them. The surface area is too small and the stakes too high for current models to cover. The **continuity doctor** survives, the long-term physician who knows your history, your family, your context, and provides judgment that weaves together everything an AI might generate. This is closer to the old "family doctor" model than what most modern GPs actually do. Not screen-reading appointments. Real ongoing care. The **interventional generalist** survives, the doctor who does the physical exam, runs the procedure, performs the minor surgery, takes the biopsy, sets the bone. Hands-on work that AI literally cannot do without a robot, and we are nowhere near robots that can do this work outside of narrow surgical contexts. What dies is the **referral-machine GP**. The person whose job is to listen to symptoms, pattern-match against a decision tree, and forward you to the right specialist. That role is being replaced by a phone with a camera and an AI app. It's not a question of if. It's a question of how fast the regulatory bodies and medical associations can slow it down before it happens anyway. If you're a patient, this is mostly good news. You will, within a few years, have access to specialist-level diagnostic capability in your pocket, 24 hours a day, for free or nearly free. You will spend less time waiting. You will catch problems earlier. You will not be at the mercy of a GP's bad day or a clinic's understaffing. You will still need real doctors for procedures, surgery, hospitalization, and complex multi-system disease. But the bottleneck, the GP gatekeeper, is going to crumble. And good riddance. If you're a doctor, you need to think hard about which of the three surviving categories you're going to be in. The generalist primary care role is the most exposed. If you're a young doctor choosing a specialty, this matters. If you're a mid-career GP, you need to start building toward something that AI doesn't make redundant. If you're a regulator or medical association president, the temptation will be to slow this down. To define "practicing medicine" so broadly that AI tools fall under your jurisdiction. To require human signoff on everything. To require licensed physicians to be in the loop. Some of this is reasonable. Most of it will be obstruction in service of protecting incumbent rents, and it will fail in the same way the taxi medallion system failed against Uber. The economics are too lopsided. People will route around it. If primary care becomes AI-mediated, the human relationship piece of medicine gets harder, not easier. Doctors today are already strained, fragmented, and burned out. Replace the GP with an app and we may end up with a medical system that's diagnostically more accurate but emotionally even more sterile. The person with a chronic illness who needed someone to listen as much as they needed a diagnosis loses the human entirely. The good version of this transition addresses that explicitly. AI handles the diagnostic and routine work. Humans get redeployed toward what humans are uniquely good at: presence, listening, building trust, navigating complex emotional and physical care. The doctor becomes more like a coach and less like a switchboard. The bad version of this transition just removes the human entirely to save money, and we end up with healthcare that's technically excellent and emotionally dead. Which one we get depends on how we structure it. The technology is neutral. The deployment is a policy choice. Medicine is the second-oldest guild after the priesthood, and it has been protected by the strongest combination of regulation, prestige, and information asymmetry of any profession in history. That protection is now hitting a tool that doesn't care about prestige and doesn't respect regulation by jurisdiction. The patient with a rash will use the app, regardless of whether the medical association approves. The young person with anxiety will use the AI therapist, regardless of whether the psychiatric establishment recognizes it. The parent with a sick child at 2 AM will get a better answer from a phone than from an overworked telehealth resident. The role of the doctor will narrow. The role of the specialist will expand. The role of the GP referral-machine will end. The patient gets a better deal. The healthcare system gets cheaper and faster. The losers are the people whose entire career was built on being a human bottleneck in a system that doesn't need bottlenecks anymore. There will be resistance. There always is. But the patient with a rash isn't going to wait two weeks for a fifteen-minute appointment when their phone can give them a 91%-accurate answer right now. The end of the generalist isn't a tragedy. It's an upgrade. --- # The End of Lawyers URL: https://www.askgib.com/articles/the-end-of-lawyers Published: 2026-03-23 Reading time: 6 min > Most legal work is pattern matching across precedent. AI does pattern matching better than humans, faster, cheaper, and 24 hours a day. The billable hour is finished. You hire a lawyer for a contract review. They charge you $500 an hour. You watch them open the document. You watch them go through it. They circle a clause. They write a comment. They flag another clause. They write another comment. By hour three, they've identified eleven issues, none of which surprise either of you. The contract is mostly boilerplate. The "review" is mostly verification that the boilerplate matches what boilerplate is supposed to look like. The bill arrives. $1,500. This entire transaction is now a $30 problem solved by software. The mystique of the legal profession is built on the idea that law requires deep, specialized human judgment. And in *some* cases, it does. A complex international M and A negotiation. A novel constitutional argument before a supreme court. A bet-the-company patent dispute. These require lawyers who can think creatively, build narratives, and read rooms. | What lawyers actually spend their time on | Share | |---|---| | Document review / e-discovery | ~25% | | Legal research | ~20% | | Contract drafting / review | ~20% | | Client meetings / communication | ~15% | | Court appearances / depositions | ~10% | | Administrative / billing | ~10% | *The first three categories, roughly 65% of all legal work, are precisely what AI is now doing better, faster, and cheaper. The remaining 35% is what survives.* The bulk of legal work is pattern matching. Read a contract, compare it to thousands of similar contracts, flag the deviations. Research case law on a specific question, summarize the precedent. Draft a will. Draft a lease. Draft an employment agreement. Review discovery documents and tag the ones relevant to the case. File compliance paperwork. Send a demand letter from a template. This is not creative legal thinking. This is structured information processing at scale. The reason it costs $500 an hour is the artificial scarcity of people licensed to do it, not the cognitive difficulty of the work itself. Thomson Reuters acquired Casetext in 2023 and integrated their AI legal assistant, CoCounsel, into Westlaw, the dominant legal research platform in North America. By 2025, CoCounsel was being used by major law firms including Morgan Lewis and Polsinelli. The system can read entire case files, run multi-step research plans, draft complaints, generate employee policies, and produce first drafts of complex filings like SEC Form 8-K reports. Thomson Reuters' own data says the system delivers **2.6x faster document review and contract drafting**, with **85% of users finding more key information** than they did manually. Their research projects that widespread AI adoption will free up **about 200 hours per year per lawyer**, or roughly 10% of total billable capacity. | | Hours per year | |---|---| | Manual legal research | ~500 | | AI-assisted legal research | ~300 | | **Savings** | **~200 hours / lawyer / year** | *Thomson Reuters' own 2025 data. This is the conservative projection, assuming lawyers continue to do most work with AI assistance. The aggressive scenario replaces three lawyers with one.* That projection is conservative. It assumes lawyers continue to do most of their work and use AI to assist. The more likely outcome is that one lawyer with AI replaces three lawyers without. The reduction isn't 10%. It's 60-70%. Harvey AI, used by major firms including A&O Shearman, is doing the same thing. So is Lex Machina, which provides predictive analytics on judges, courts, and counsel. Spellbook, for contract review. Everlaw, for e-discovery. The entire stack of legal work below the partner level is being absorbed into software. And here's the part that should worry the profession most: courts are now using these tools too. **CoCounsel is being marketed directly to judges.** When the judge has the same AI tool as both sides' lawyers, the bottleneck moves to the parts of the case where humans actually add something. Which, increasingly, is a small fraction of the total work. The billable hour exists because it's the only way to charge a client for time without committing to an outcome. Doctors don't bill by the minute. Engineers don't bill by the screw turned. Lawyers bill by the hour because hourly billing perfectly aligns the lawyer's incentive with *not finishing*. If you're billed by the hour, your lawyer is paid more when the case drags on. They're paid more when discovery is exhaustive. They're paid more when there are six rounds of revisions. The legal industry has built an entire business model on rewarding inefficiency. When AI compresses a six-hour task into thirty minutes, the firm has two choices. Pass the savings to the client, or bill the same six hours and pocket the difference. Most firms will try the second strategy. It will work for a few years. Then competition from AI-native legal services, at one-tenth the cost, will gut the market. The smart firms know this is coming. They're rebuilding around fixed fees for routine work. The dumb firms are still pretending the billable hour will last forever. Both groups will be smaller in five years. The **specialist**, a lawyer with deep expertise in a narrow, high-stakes domain. Patent litigation in semiconductors. Cross-border tax in three specific jurisdictions. Constitutional law at the appellate level. Niches where AI helps but doesn't replace, because the surface area is too small for general-purpose models to cover, and the consequences of being wrong are too large to delegate. The **negotiator**, a lawyer who is really a strategist with a JD. Someone who reads people, builds coalitions, and closes deals. M and A advisors. White-collar defense lawyers who manage corporate clients through investigations. Litigators who try cases. None of this is going to be done by software, because the substance is human relationships, not legal reasoning. The **builder**, a lawyer who runs an AI-native law firm. Someone who built the systems that let one person do the work of fifteen. These firms don't look like Cravath. They look like software companies that happen to provide legal services. They will, over time, absorb the routine work of the entire industry. Everyone else, the associate doing document review, the contract attorney, the routine commercial lawyer, the family law generalist working from a template, the immigration paralegal, is being directly competed against by software that's already better, already cheaper, and improving faster than they are. If you're a person who occasionally needs legal help, a contract reviewed, a will written, a small claim, a divorce, a real estate closing, your options are about to get much better. In 2026, you can already use ChatGPT or Claude to review a basic contract and flag the unusual clauses. It will catch most of what a junior associate would catch, at zero cost. For anything more serious, AI-native legal services are emerging that charge a flat fee, $300 to $1,500, for work that used to cost $5,000 to $15,000. The legal profession will resist this. Bar associations will lobby for "unauthorized practice of law" enforcement against AI tools. Some of those efforts will succeed in the short term. None will succeed for long. The economics are too lopsided. The information asymmetry that justified the price for the last hundred years is gone. For corporate clients, the same logic applies at scale. A general counsel with a small in-house team and AI tools can now do what required a 20-person legal department a decade ago. The big firms still get the bet-the-company work. They lose almost everything else. There's one real concern that the AI optimists tend to gloss over. Law isn't just about getting the answer right. It's about getting the answer right *in a way that holds up to scrutiny*, in a system where the consequences of being wrong include people going to prison or losing their homes. AI systems still hallucinate. They still make confident errors. They still miss context that a human lawyer would catch. The famous case of a New York attorney who submitted a brief containing AI-fabricated case citations in 2023 was a warning shot. There have been others since. The right response to this isn't "AI can't be used for law." It's "AI must be used carefully, with verification, and within clearly defined boundaries." Which is what good lawyers are already doing. CoCounsel is built around grounding every answer in verified Westlaw content, with citations linking directly to source material. The verification problem is solvable. It just requires the humans to actually verify. The remaining role of the lawyer, increasingly, is to be the verifier. To be the human who reads the AI output, checks it against reality, and signs the final document. That role is real. But it's one person where there used to be ten. For two thousand years, the law has been a guild. Membership was restricted. Knowledge was kept obscure. The barrier to entry justified the price. In the next ten years, that ends. The knowledge is no longer obscure. AI has memorized every case, every statute, every standard form, every precedent. It's faster than a human, cheaper than a human, and improving while humans stagnate. The lawyers who adapt, who become specialists, negotiators, or builders, will do fine. The ones who don't, the army of generalists doing pattern-matching work at $300 to $700 an hour, will discover that their entire job has been quietly absorbed into a chatbot. The age of legal services as an artisanal craft is ending. What replaces it is faster, cheaper, more accessible, and, for the routine cases that make up most of the work, usually just as good. The billable hour was always a bug in the system. AI is the patch. --- # Canadian Healthcare Eats the Budget URL: https://www.askgib.com/articles/canadian-healthcare-eats-the-budget Published: 2026-03-16 Reading time: 2 min > Canada will spend a record $399 billion on health care this year, and the median wait still hit 28.6 weeks. More money, longer lines, every single year, in a system that won't even count who it pays for. Canada will spend about $399 billion on health care in 2025. That is a record. It is roughly 12.7 percent of everything the country produces, and it is growing faster than the economy itself. The median wait from a family doctor's referral to actual treatment is now 28.6 weeks. More than half a year. In 1993 it was nine. More money. Longer lines. Every single year. This is not a small budget or a shrinking one. Health care is the single biggest line item in every province, and it grows faster than everything else those governments spend on. Hospitals take the largest slice, about a quarter of the total, and inside hospitals roughly two thirds of the cost is staff pay, which jumped 13 percent in one year. Doctors are another 14 percent or so, drugs another 13. Enormous inputs. And yet about 6 million Canadians have no family doctor, and an MRI is an 18 week wait. So where does the money leak. Start small. I broke my ankle, and the hospital tried to charge me $250 plus tax for a cast. I found the same cast on Amazon for $100. From the source it was $20. A system that buys for 40 million people has the purchasing power to pay $20, and it pays retail plus because the money passes through a dozen hands before it reaches the cast. Now multiply that by every glove, every bandage, every bedpan in the country. The system also pays for things working citizens don't get. Refugee claimants receive dental, vision and physiotherapy through a federal program, coverage an employed Canadian doesn't have. Even after Ottawa added co-pays in 2026, a claimant pays 30 percent of a dental bill while an uninsured citizen pays all of it. And there is a hole nobody will fill. No government will tell you how much of the health budget goes to people who are here temporarily or without status, because no one counts it. Ottawa's own figures show roughly 1.4 million temporary residents with expiring permits in 2026 alone, on top of 1.5 million that expired in 2025, students not counted, and that number resets and stacks year after year. Many won't leave. A system this expensive that chooses not to measure who it pays for, that is the real scandal, bigger than any single figure. And no, the doctors are not the problem. We don't have bad doctors, we have too few of them, and then we make qualified foreign-trained doctors re-train for years while patients sit on a list. We ration the supply ourselves, then act surprised at the wait. Here is the part that should end the argument. Spending more has not worked, and it cannot work in a system that gets less efficient the bigger it grows. Australia and the Netherlands buy capacity from private hospitals to clear the queue, and they get better outcomes per dollar. We treat that idea like heresy, so the line stays where it is. We keep writing bigger cheques to the same machine and calling it care. The cheque clears. The line does not move. --- # Your Taxes Should Build Your Neighborhood URL: https://www.askgib.com/articles/your-taxes-should-build-your-neighborhood Published: 2026-03-09 Reading time: 5 min > You pay 50% of your income to a government that builds roads in places you'll never see. Here's a different model. You pay roughly half your income in taxes. Income tax. Sales tax. Property tax. Payroll tax. Capital gains tax. Sin tax. Carbon tax. Every layer of government takes its cut. You add it all up and the number is staggering. In most developed countries, a middle-class earner sends 40 to 55% of their lifetime income to the state. Not the country. Not the abstract idea of "public services." You. Your block. Your neighborhood. The school your kids actually attend. The hospital closest to your house. The library you'd actually use. For most people, the honest answer is: very little that's visible. Tax money flows through a black box. You pay into the top. Decisions get made somewhere far from you. Money comes out the bottom, building a road in a region you'll never visit, funding a study you'll never read, paying a contractor who happens to know someone in government, subsidizing an industry whose lobbyists wrote the law. Some of it does come back to you. Public schools exist. Roads exist. Healthcare, in countries that have it, exists. But the allocation isn't transparent, and the link between your contribution and the services around you is almost completely severed. | Where your tax dollar goes (typical developed country) | Share | |---|---| | Pensions / social security | ~25% | | Healthcare | ~20% | | Education | ~12% | | Civil service / administration | ~10% | | Debt servicing | ~8–12% | | Defense | ~7–10% | | Infrastructure (national) | ~5–8% | | Subsidies & transfers to industry | ~5–7% | | Other / discretionary | ~5–10% | | Foreign aid / international | ~1–2% | *The categories sound reasonable. The problem is that almost none of it is visible to the person paying. Most of what you contribute disappears into national-level allocations that bear no relationship to your daily life.* This is the core problem. You pay enormous sums into a system whose outputs you can't see, can't measure, and can't influence. So you assume it's working, or you assume it's not, depending on your politics. Both sides are mostly guessing. What if a portion of your taxes, say, 20 to 30%, was earmarked for direct, local infrastructure that you can see? Your neighborhood. Your zip code. Your block. The school three streets over. The clinic in your district. The library at the corner. The transit stop two blocks down. The hospital you'd actually go to in an emergency. Not roads in some region you'll never visit. Not subsidies for industries that don't employ anyone you know. Not contractors who happen to be friends with whoever sits in your provincial legislature. The argument against this model is that it makes rich neighborhoods richer and poor neighborhoods poorer. That argument deserves serious examination, but here's why it doesn't actually hold up. People in poor neighborhoods today are already getting nothing. Their tax money is going to the same black box. The current system doesn't redistribute their taxes into better services for them, it redistributes their taxes into political projects, contracts, and overhead. The poor are paying in, just like everyone else, and getting back roughly the same: a fraction of what they paid, plus a lot of administrative noise. Under a model where taxes built your actual neighborhood, the dynamic is different. A poor neighborhood with 5,000 residents earning median wages still generates millions in tax revenue every year. That money, instead of disappearing, would visibly improve the place where those 5,000 people actually live. The library upgrades. The clinic gets a second doctor. The school gets new equipment. Modest, but visible. And accumulating year after year. The richer neighborhood, with higher incomes and higher tax contributions, would obviously build faster. That's true. But it's the dynamic that's been at work in the housing market for fifty years anyway, and right now, the rich neighborhood is getting its services *and* harvesting the tax revenue of the poor neighborhood through the central budget. Under the proposed model, at least the poor neighborhood keeps what it generates. Swiss federalism pushes most taxation and spending decisions down to the canton level, and within the canton, down to the commune. There are 26 cantons and over 2,000 communes, many of them small enough to know what their residents actually need. A Swiss citizen pays federal tax, but also pays cantonal tax and communal tax, and the bulk of what they pay funds services in places they can see. *Both models collect roughly the same amount. The difference is whether you can see what you paid for.* The result is one of the highest-functioning public infrastructure systems in the developed world. Swiss trains run on time. Swiss hospitals are exceptional. Swiss public spaces are well-maintained. And, critically, Swiss citizens vote directly on major spending decisions, multiple times a year. If the commune wants to build a new pool, the residents vote. If the canton wants to fund a hospital expansion, the residents vote. This isn't a perfect system. Switzerland has its own problems. But the principle works: taxes paid locally, decisions made locally, results visible locally. The accountability loop is short enough that you can see, in your daily life, what your money is doing. Most other developed countries took the opposite path. Decisions centralized at the national level. Spending decided in capitals far from the people paying. Tax money disappearing into ministries staffed by people who will never set foot in the neighborhoods they're allocating money to. If your taxes build your neighborhood, you have a reason to pay them. You have a reason to want a better job. You have a reason to move into a better area, because moving into a better area means your taxes start building that area too, and the area gets even better. Your contribution becomes visible. The link between effort and reward is restored. Under the current system, that link is broken. You work harder, you earn more, you pay more in tax, and almost none of it visibly improves the place you actually live. So why work harder? Why move? Why invest in your community? Your money is going to a federal program you don't understand, run by people you didn't elect, benefiting communities you'll never visit. People are responding to incentives. The current incentive structure says: your taxes are a price you pay for permission to live in a country, not an investment in the place you actually live. That's a soul-crushing structure. And it's the source of a lot of the political resentment we see in developed countries right now. A model that ties taxes to visible local outcomes would change that. People would want to contribute more if they could see what they were getting. People who refused to contribute would get fewer local services, which is fair. People who moved to a struggling neighborhood would be moving with the expectation that their contribution would help fix it, which is exactly what we want. The federal government keeps a portion of tax revenue for genuinely national functions, defense, foreign policy, interstate transport, the courts, monetary policy. Maybe 40 to 50% of total tax intake. The rest is allocated locally. The local share is split between the city level and the neighborhood level. Each neighborhood (defined as something like a school catchment area, or a postal code) has a budget that matches the tax revenue generated by its residents. That budget funds the schools, clinics, library, parks, and small infrastructure in that neighborhood. Residents vote on how to allocate the neighborhood budget, several times a year. New playground or better staffing at the clinic? The neighborhood decides. Repaving the road or hiring more teachers? The neighborhood decides. The vote is direct, the budget is small enough to be meaningful, and the result is visible within a year or two. A neighborhood that consistently votes for short-term spending and lets its infrastructure decay will have visibly worse infrastructure. A neighborhood that invests in its schools and clinics will visibly improve. People will see this and move accordingly. Cities will start competing on the quality of their neighborhoods, the way they currently compete on tax rates. You pay roughly half your income to the state every year of your working life. You should be able to point at something you can see and say: that's where it went. Not abstract national priorities. Not roads in the middle of nowhere. Not contracts to friends of the government. Not a study commissioned in 2014 that's still being written. A library. A clinic. A bus stop. A school. Something you walk past on the way to work and know your contribution helped build. This is not a radical idea. It's how municipalities used to work before the centralization wave of the 20th century. It's how Switzerland still works. It's how most people, if asked plainly, would say they want their money spent. That's how a tax becomes an investment instead of a tribute. --- # Insurance Is a Scam URL: https://www.askgib.com/articles/insurance-is-a-scam Published: 2026-03-02 Reading time: 6 min > A 4,000-year-old industry built on collecting money and refusing to pay it back. The pyramid is real, and Warren Buffett built his fortune on it. You pay insurance every month for a decade. You never make a claim. You're a "good customer." Then something happens. You make a claim. Suddenly you're treated like a fraud suspect. Your file goes to an adjuster whose job performance is partly measured on how much money the company *doesn't* pay out. They request documents. They request more documents. They cite clauses you've never read in a contract you didn't write. They lowball the offer. They drag the process for months. You either accept 40% of what you're owed, or you hire a lawyer and fight for two years. That's insurance. That's the product. The concept of insurance is ancient. The Code of Hammurabi, written around 1750 BC in Babylon, contains the first documented form of it. Merchants would take a loan to fund a sea voyage. If the ship sank or the cargo was stolen by pirates, the loan didn't have to be repaid. The interest covered the risk. This was called "bottomry," and it's the seed of every modern insurance contract. Code of Hammurabi documents bottomry (Babylon) Risk pooling itself is older than that. Chinese river traders, around 3000 BC, would split their cargo across multiple boats so that one shipwreck wouldn't ruin them. The Romans had burial societies where members paid small monthly dues into a pot that covered funeral costs when one of them died. Alpine farmers helped each other when livestock or children fell ill. All of it was insurance, in the original meaning of the word: a group of people pooling money so that no individual would be wiped out by an unlikely catastrophe. Hindu and later Greek traders adopt bottomry Roman burial societies, first life/funeral insurance First documented marine policy in Genoa Pedro de Santarém publishes first book on insurance Great Fire of London, Nicholas Barbon founds first fire insurance company The structure that runs the modern world started in a London coffeehouse in 1688. Edward Lloyd's coffee shop became a meeting point for merchants, ship owners, and investors. When someone needed insurance for a voyage, the cargo manifest was passed around. Investors who wanted a piece would sign their names underneath the document, agreeing to cover a percentage of the loss. That signature is where the word "underwriter" comes from. Lloyd's of London still exists. It's now a $50+ billion market. Edward Lloyd's coffeehouse opens, Lloyd's of London is born Lloyd's formalizes as an underwriting market Modern publicly-traded insurance corporations emerge Buffett takes over Berkshire Hathaway, buys its first insurer in 1967 and starts the float strategy Global market 7 trillion USD premiums, reinsurance 500 billion USD The original idea was beautiful. The modern version is a very different animal. Original insurance was mutual. The people paying in were also the people getting paid out. The pool existed to serve them. Any surplus went back into the pool or was returned to members. Modern insurance is a publicly traded company with shareholders. The pool exists to generate profit. Every claim is a cost. Every premium is revenue. The math of the business is simple: collect as much as you can, pay out as little as you can, invest the difference. This is called the "float," and it's the entire game. The insurance company holds your money for the years between when you pay your premium and when (if ever) they pay a claim. During those years, they invest your money. They keep the investment returns. You get nothing. Warren Buffett, who is one of the richest people in the world with a net worth of around **$152 billion**, built his fortune almost entirely on insurance float. Berkshire Hathaway's insurance subsidiaries (GEICO, General Re, BH Reinsurance) generate roughly **a quarter of Berkshire's total revenue** and provide the cheap, long-duration capital that funded every other Berkshire investment for sixty years. Buffett figured out, before most people, that the actual insurance business doesn't have to be profitable as long as the float is large and stable. The insurance is the marketing. The investment income is the business. Sound familiar? It's the same model as airlines and their frequent-flyer programs. The visible product is the bait. The financial mechanism behind it is the actual money-maker. Here's the modern reality, in numbers from 2023. US health insurance plans sold through the federal marketplace denied **19% of in-network claims**. Out-of-network claims were denied at **37%**. UnitedHealthcare and AvMed, two of the worst offenders, hit denial rates of **33%**. *Source: KFF (Kaiser Family Foundation) 2023 ACA Marketplace data; CMS reporting. One in five medical claims, on a plan you're already paying for, gets denied on the first pass. The worst insurers deny one in three.* The reasons are mostly opaque. The most common category for denial in 2023 was literally labeled "Other", meaning the insurer didn't specify a reason. The second most common was "administrative." A small fraction were denied for actual medical reasons. Auto and property insurance work the same way. You pay your premium for years without issue. The day you have an accident, the company sends an adjuster whose job is to find every possible reason the claim shouldn't be paid in full. Pre-existing damage. Improper documentation. Policy language. Delay in reporting. They lowball. You fight. Most people give up. And here's the obscene part: insurance companies know what happens when people fight back. In Medicare Advantage, insurers reverse **more than 80% of the treatment denials that get appealed.** On the marketplace, fewer than 1 in 100 denied claims ever gets appealed. They deny first, knowing most people won't fight. The denial isn't a judgment about the claim. It's a filter. A way to keep money in the float by exhausting the customer. The really wild part is that insurance companies are themselves insured. By other insurance companies. It's called reinsurance, and the global market is worth around **$500 billion a year**, projected to hit $1 trillion by 2035. The top five reinsurers, Munich Re, Swiss Re, Hannover Re, Berkshire Hathaway Reinsurance, and SCOR, control about 40% of the market between them. *Each layer collects fees and holds float. When the bottom layer needs to make a claim, every contractual reason not to pay sits ready in the layers above.* And the reinsurers themselves use retrocession, they re-insure their own reinsurance. The risk gets diced and spread across institutions until no one really knows who's exposed to what. This isn't speculation. This is how the global insurance system is structured. Each layer collects fees. Each layer holds float. Each layer earns investment income on money that originated, somewhere at the bottom of the pyramid, with a customer paying a monthly premium. When something goes truly catastrophic, a 2008-style financial crisis, a pandemic, a war, the bottom of the pyramid pays. Customers can't get claims paid because the primary insurer says it's a force majeure. The primary insurer turns to the reinsurer, who turns to the retro-reinsurer, who turns to the capital markets. Everyone has a contractual reason not to pay. The losses get pushed back down to the customer, the taxpayer, or both. Look at what happened with cyber insurance during the major ransomware waves of the early 2020s. Companies that paid for years of cyber coverage discovered, when actually attacked, that their policies excluded "acts of war," which insurers then defined to include state-sponsored hacking groups. Coverage that customers thought they had, denied. The premiums had already been collected and invested. The losses got dumped back on the victims. The scam works because in most of the developed world, insurance is mandatory. You can't legally drive a car in any developed country without auto insurance. You can't get a mortgage without home insurance. You can't operate a business without liability insurance. In the US, until 2019, you literally couldn't *exist* without health insurance, the individual mandate penalized you for not buying it. This is a hostage market. You don't get to opt out. You don't get to negotiate. You take what's offered, you pay what's asked, and you hope you never actually have to use it. Because the moment you try to use it, the company you've been paying for years transforms from a friendly brand into a legal adversary. And the regulators? Most insurance regulation is done at the state level in the US and the national level in Europe, by regulators heavily influenced by the industry they oversee. The revolving door between insurance executives and insurance commissioners is well-documented. The result is decades of regulatory capture, with consumer protections that look strong on paper and fail in practice. Insurance, in its original form, was a beautiful idea. A group of people pool small amounts of money so that when one of them faces a catastrophe, the group covers it. Risk shared, individuals protected. This still exists, in pockets. Mutual insurance companies (companies owned by their policyholders, not shareholders) operate on the original model. Cooperative insurance, common in parts of Europe, returns surplus to members. Some peer-to-peer insurance startups have tried to rebuild the model from scratch, with mixed results. But the dominant model, publicly traded insurance corporations with quarterly earnings targets and shareholders who want returns, has structural incentives that point in exactly the wrong direction. Their job is not to pay claims. Their job is to maximize the float, deny what they can, and return profit to shareholders. The original insurance was a community protecting each other. The modern version is a financial institution using your premiums as raw material for capital markets, occasionally interrupted by the inconvenience of having to actually pay out a claim. You don't have insurance. You have a subscription to a denial machine. When you finally try to use the product you've been paying for, you discover the company has spent the last decade quietly preparing legal and procedural defenses against you. The premium was the easy part. The claim is the war. Warren Buffett figured out that this entire industry isn't really about protecting people. It's about collecting capital from millions of small payers, deploying it at scale in financial markets, and counting on enough customers giving up before they get paid out. Insurance is mandatory. Insurance is profitable. Insurance is a four-thousand-year-old concept that got corrupted somewhere between Hammurabi and the IPO. You can't escape it. You can only understand what it actually is. A financial product, sold as protection, owned by people who win when you lose. --- # The Death of Consultants URL: https://www.askgib.com/articles/death-of-consultants Published: 2026-02-23 Reading time: 6 min > A $500 billion industry built on copy-paste, by their own admission, fails 70% of the time. AI is going to finish the job. You hire a consulting firm. They send three twenty-six-year-olds with MBAs. The twenty-six-year-olds spend two weeks interviewing your staff, asking questions your staff have been answering in internal meetings for years. They build a 90-slide PowerPoint. The slides look identical to the ones the same firm built for a competitor last quarter. The recommendations are obvious. The price is $1.2 million. You implement 30% of it. It doesn't really work. The firm offers to come back next year to "refine the rollout." Big consulting firms charge a billion-dollar premium for what is essentially a templating service. They build a deck once, refine it across thousands of engagements, and resell it to every new client with the variable names changed. The actual thinking happened at scale, in a partner's head five years ago. The associate isn't paid to think. They plug your numbers into the template, polish the slides, and translate the partner's hand-waving into bullet points. You don't pay $500 an hour for a template. You pay $500 an hour because the template is invisible. The illusion is that someone smart spent months tailoring a custom solution to your unique problem. The reality is the solution was custom in 1998, and you're buying its 800th iteration. Companies pay this premium for two reasons that have nothing to do with the advice. To defer risk (if the strategy fails, "well, McKinsey said to do it"). And for deductibility (consulting is an operating expense; a senior hire would be a long-term commitment with severance, benefits, social charges, and a real legal relationship). The advice is incidental. The firm is renting you cover. The global management consulting industry generates somewhere between **$370 billion and $1 trillion a year**, depending on how you count it. Roughly **3 million businesses** are involved. North America alone accounts for over $150 billion. For perspective: that's larger than the global music industry, the global film industry, and the global video game industry combined. For a service whose primary deliverable is a PowerPoint. McKinsey, BCG, and their peers publish research on how often corporate transformations succeed. Their own answer: **70% fail.** Out of 100 transformation projects: - ~70 fail outright - ~30 succeed - Of those 30, only ~25% (~7-8) sustain the gains - Final tally: **roughly 7 out of 100 actually work.** *McKinsey, BCG, and others publish these numbers themselves. Seventy percent of transformations fail. Of the thirty that succeed, only a quarter sustain the gains. Out of 100 projects, roughly seven actually work.* The firms whose entire business is selling transformation projects publicly state that 70-95% of those projects don't work. And they continue charging seven-figure fees for them. If a surgeon failed 70% of their operations, they'd be in prison. If a contractor's houses collapsed 70% of the time, they'd be bankrupt. In consulting, you publish the failure rate as a research insight and use it to sell the next engagement. The selling point is: "Most transformations fail. That's why you need us." A bit like a roofer telling you most roofs leak, which is why you should hire him to install another one. Most consulting failures are boring. The deck collects dust on a shared drive, the company quietly pivots, the consultants move on. But occasionally a project lands somewhere catastrophic. | Year | Firm | Scandal | Penalty / Damage | |---|---|---|---| | 2001 | Arthur Andersen | Enron audit fraud | Firm collapsed; $60B+ Enron bankruptcy | | 2002 | Arthur Andersen | WorldCom | $11B fraud; 30,000 jobs lost | | 2016–17 | McKinsey | South Africa / Eskom / Gupta | $74M returned; severe damage to SA economy | | 2021 | McKinsey | Purdue / opioid crisis (state settlements) | $573M across 47 US states | | 2024 | McKinsey | Purdue / opioid crisis (federal DOJ) | $650M; criminal felony obstruction | | 2024 | Publicis Health | Opioid advisory work | $350M settlement | *Not a complete inventory. Just the cases big enough to make the news.* **Arthur Andersen and Enron (2001).** Andersen was one of the Big Five, serving as Enron's auditor and consultant simultaneously, billing about $1 million a week split nearly evenly between the two. They didn't just miss the irregularities, they helped engineer them. When the SEC came knocking, Andersen ordered staff to shred documents. The firm collapsed. Enron went bankrupt with over $60 billion in assets gone, taking pensions and savings with it. The consequence for the consulting industry as a whole? Roughly zero. The Big Five became the Big Four. Business continued. **McKinsey and the opioid epidemic.** Between 2004 and 2019, McKinsey advised Purdue Pharma on how to "turbocharge" the sales of OxyContin. They built playbooks for targeting high-prescribing doctors. They proposed rebate programs to pharmacies for overdoses. When the federal investigation began, a senior McKinsey partner personally deleted relevant documents. McKinsey eventually paid **$650 million** to the US Department of Justice in 2024, on top of nearly **$900 million** in earlier state settlements. The opioid epidemic has killed over **600,000 Americans** since 1999. No McKinsey partner has gone to prison. The firm continues to operate, including on government contracts. **McKinsey and South Africa's state capture.** McKinsey partnered with a shell company linked to the Gupta family to win a $120 million contract with Eskom, South Africa's state-owned power utility. The Zondo Commission later found "rampant corruption" tied to the transactions. McKinsey eventually returned about $74 million in fees and apologized publicly. The damage to South Africa's energy infrastructure and public finances is still being calculated, years later. These aren't bad apples. These are systemic patterns: when a consulting firm's incentive is to win the next engagement, oversight becomes optional. The client is whoever pays. The ethics are negotiable. Consulting isn't entirely worthless. Operational turnarounds in distressed companies sometimes need an outside hand, because internal leadership is too compromised by politics to make hard cuts. Specialized expertise, things like pricing strategy, supply chain optimization, regulatory compliance in a new market, is real value when the firm has actual deep knowledge. M and A integration occasionally benefits from a dispassionate third party. And the original wave of strategy consulting (think Bruce Henderson at BCG inventing the growth-share matrix in the 1970s) really did introduce frameworks that didn't exist before. But those cases are a small fraction of what gets sold. The bulk of the industry is selling generic "transformation" decks to companies that already know what's wrong and don't have the political will to fix it. The cleaner version exists: small boutique firms with real domain experts, independent advisors who built and sold companies, specialists who actually do the work. None of them charge $500K for a 90-slide deck made by people fresh out of school. The McKinsey model depends on three things: the templates, the polish, and the gravitas of the brand. AI demolishes the first two and exposes the third. A mid-level executive with access to a good language model can now produce a strategy deck in three days that's structurally indistinguishable from a McKinsey deliverable. The frameworks are public. The "proprietary methodologies" were always just rebranded Porter, Drucker, and Christensen. AI compresses what an associate took six weeks to do into an afternoon, and increasingly does it better, because it doesn't make formatting errors at 2 AM. The consulting firms know this. McKinsey deployed an internal AI tool called **Lilli** that, by mid-2025, was being used by 70% of their consultants to automate proposals and presentations. The people you're paying $500 an hour are themselves using AI to do the work you're paying them to do. Companies are catching on. Boards are receiving million-dollar strategy presentations that turn out to be lightly edited LLM output. Once you've seen it, you can't unsee it. The premium evaporates. A company with fifty years of operational experience does not need a twenty-six-year-old to tell it what to do based on a ChatGPT prompt. The information asymmetry that justified the price is gone. | $1.5M → Big Consulting Firm | $1.5M → In-House Over 5 Years | |---|---| | 6-month engagement | 1 senior strategist on $300K/year (or 2 specialists at $150K) | | 3 associates + 1 partner | Lives the strategy daily | | 1 PowerPoint deck (~90 slides) | Knows the company history | | 0 implementation | Can be fired if they fail | | 0 ongoing accountability | Skin in the game | | ~7% probability of lasting impact | Cumulative knowledge stays in the company | *Same money. Different outcomes. The math isn't close. The reason companies still pick the consulting deck has nothing to do with results, it's the deniability that comes with it.* The reason companies don't do this isn't economic. It's psychological. Hiring requires the executive making the hire to take responsibility for the outcome. Consulting is rented expertise: a contract, a deliverable, a clean exit. When the strategy fails, no one gets fired. The deck goes in a drawer. Everyone moves on. This is the actual product consulting sells. Not strategy. Not insight. Plausible deniability for executives who don't want to make decisions themselves. It's a $500 billion business that fails 70% of the time by its own measurement. It funded the opioid epidemic. It helped engineer the Enron collapse. It enabled state capture in South Africa. It has been quietly extracting fees from companies that needed coaches, not consultants, for fifty years. And now AI can do most of what they sell, faster, cheaper, and without the conflicts of interest. The death of consultants isn't a prediction. It's a phase transition. The firms that survive will be the small specialists who actually know something. The rest will quietly shrink into legacy brands selling nostalgia to executives who haven't realized the joke is on them. Pay your people. Hire experts. Take responsibility for your own decisions. The age of the deck is over. --- # Pay Your Staff URL: https://www.askgib.com/articles/pay-your-staff Published: 2026-02-16 Reading time: 6 min > Tipping used to be 5%. Now it's 30%. Same plate, same effort, three times the guilt. This is what happened. You finish dinner. The iPad spins around. 18%. 22%. 25%. Custom. There's no zero option. There's no 10. The lowest button is already 18, and the person who handed you the iPad is standing two feet away, watching. You're on a date. Or with your boss. Or with friends. So you tap 22, because tapping 15 in front of another human being now feels like an act of war. That's not a tip. That's a tax on social discomfort. Go back forty years. In North America, a tip was 5% for normal service. 10% for exceptional. That was the deal. It was a small thank-you on top of a meal that was already priced to pay the staff. Then it crept. 10% became standard. Then 15%. Then 18%. Now the prompt screen starts at 20% and the "good service" tier is 30%. Nobody voted on this. It just happened. And it happened because restaurants figured out they could shift their entire labor cost onto the customer and call it gratitude. | Year | Standard tip | Good service | |---|---|---| | 1950 | 5% | 10% | | 1970 | 10% | 12% | | 1990 | 12% | 15% | | 2000 | 15% | 18% | | 2010 | 15% | 20% | | 2020 | 18% | 22% | | 2024 | 20% | 25% | | 2026 | 22% | 30% | *iPad payment terminals appeared around 2010. The "pre-set tip button" is the single biggest driver of tip inflation in modern restaurant history.* A waiter carrying a $15 burger to your table is doing the exact same work as a waiter carrying a $75 steak. Same distance. Same effort. Same tray. Same smile. But on a percentage tip, the steak waiter gets five times more for the same labor. | | Diner | Steakhouse | |---|---|---| | Meal | $15 burger | $75 steak | | Plates carried | 1 | 1 | | Time with table | ~10 min | ~10 min | | Tip at 20% | **$3** | **$15** | | Effective hourly tip rate | $18/hr | $90/hr | *Same job. 5x the pay. Because the food cost more.* It's not a service charge. It's not a skill premium. It's a tax on what you ate, paid to someone who had nothing to do with what you ate. The waiter didn't cook the steak. The waiter didn't source the steak. The waiter didn't age the steak. The waiter walked it across a room. If I work for an hour and you add $9 to my paycheck, that's a real bump. Maybe a 50% raise on the hour. Now imagine I work an hour at a steakhouse and you add $40. Same hour. Same labor. Different room. Tipping with a percentage makes no logical sense for the actual work being done. It only makes sense as a guilt mechanism. People will tell you waiters can't survive without tips because their wage is so low. In most of North America, this is no longer true. In some of it, it was never true. In Ontario, the "liquor server" subminimum wage was eliminated in 2022. Every waiter, every bartender, every food runner now makes the same minimum wage as everyone else, $17.60 an hour as of October 2025. Same as a cashier. Same as a daycare worker. Same as the kid stocking shelves at Loblaws. Seven US states (California, Alaska, Minnesota, Montana, Nevada, Oregon, Washington) require the full state minimum wage for tipped workers. No tip credit allowed. Servers in Seattle make $20+ an hour before tips. And in the states that *do* allow a tipped minimum (the federal floor is $2.13 an hour), the law still requires the employer to top up the wage to the regular minimum if tips don't cover the gap. It's right there in the FLSA. The "servers literally make nothing" framing is wrong. The legal floor is the federal minimum wage. The employer is on the hook for the difference. So when a waiter tells you they "depend on tips to survive," what they actually mean is they depend on tips to make significantly more than minimum wage. Which is what every other worker in North America wants, and doesn't get, because their boss doesn't have a mechanism to shame the customer into paying it. And here's the next thing nobody mentions: the tip you leave doesn't all go to your server. | Where your $20 tip goes | Share | |---|---| | Server keeps | $12–14 (60–70%) | | Bussers, runners | $2–3 (10–15%) | | Bartender | $2 (10%) | | Kitchen / back-of-house | $1–2 (5–10%) | | Credit card processing fee | $0.40–0.60 (2–3%) | *Most diners think the entire tip goes to the server. In most restaurants, 30 to 40% gets redistributed before the server sees a dime.* Make a list of people who do hard work for not enough money. Your kid's daycare worker. Your neighborhood pharmacist. The cashier at the grocery store. The person who delivered your couch. The mechanic who fixed your brakes. The cleaner at the hospital. The secretary at your office. The lifeguard who watched your kid swim. The bus driver. The teacher. You don't tip any of them. Nobody hands you an iPad after a doctor's appointment. Nobody puts a tip line on the bill at the hardware store. These people work just as hard. Many work harder. Many earn less. They get paid by their employer because that's how a job works. The employer figures out what the labor costs, builds it into the price, and pays the staff. Restaurants are the only industry that opted out of this system. They externalized payroll. They put it on you. And then they convinced you that *not* paying it makes *you* the bad guy. This is the part that really gets me. In most of the world, the cost of service is already in the menu price. You go to a restaurant in France or Italy or Japan, the number on the menu is the number you pay. Maybe a small rounding tip at the end if you really felt taken care of. That's it. The labor cost is built in. The price is honest. In North America, the menu price is a lie. It's the price minus 18 to 30%. Plus tax. Plus a tip you're now expected to calculate after tax, not before. Plus, increasingly, a "kitchen service fee" or "wellness fee" or "automatic gratuity" that's a tip you didn't agree to. Your $40 burger is actually $55. Charge a 15% service fee. Charge a 25% service fee. Charge a 50% service fee if your restaurant is fancy enough to justify it. I genuinely don't care. If my burger costs $75 because you've put a 250% service charge on it, then I look at the menu, decide the burger isn't worth $75, and I don't walk in. That's how prices are supposed to work. The customer sees the real number and decides. What I object to isn't the cost of running a restaurant. It's the dishonesty of pretending the cost is one thing on the menu and a different thing on the bill, and that the gap is somehow my moral responsibility to fill. Restaurants in Europe figured this out a hundred years ago. The bill is the bill. The price is the price. The waiter is paid by the boss. You eat, you pay what the menu said, you leave. Nobody's ashamed. Nobody's resentful. Nobody's quietly seething in their seat at the ratio of tax and tip and "service fee" stacked on top of a $14 cocktail. Stop tipping. Or stop the tipping system. Both work. If you're a customer, tap zero. You're not required by law to tip. You're not stealing from anyone. The waiter is being paid at least minimum wage by law. If they aren't, that's a labor violation by the *employer*, not your problem to fix at the end of the meal. If you're a restaurant owner, raise your menu prices 20% and pay your staff a real wage. That's the move. Some restaurants have already done it, Danny Meyer tried it at Union Square Hospitality Group in 2015. Around 40% of his experienced front-of-house staff left for other restaurants, and in 2020 he brought tipping back. One honest menu can't hold out alone while the place next door keeps the old trick, which is why the whole system has to go. If you're a waiter and you're reading this thinking "this guy has no idea what he's talking about, I make $60 an hour on tips and I'd never want a flat wage", fair enough. But then admit what you actually have. You have a high-paying gig that depends on a social pressure system most of your customers find uncomfortable. That's not a wage. That's a hustle. Don't dress it up as the cost of survival. And to the owners: stop hiding behind your staff. Stop using your waiters as a guilt-projection mechanism for prices you don't want to put on the menu. You're the one paying the salaries. Pay them yourself. The owner gets to advertise a $40 menu price while paying $13 an hour, knowing the difference will be made up by you, the customer, in the form of guilt translated into dollars at the moment of departure. That's the scam. Not the waiter. Not the tip. The system that lets a business owner advertise an artificially low price by quietly assigning the rest of the cost to the person eating the food. Pay your staff. Put the price on the menu. Let people choose whether your restaurant is worth it. That's not hostile. That's just honest. --- # The Beer Index URL: https://www.askgib.com/articles/beer-index Published: 2026-02-09 Reading time: 3 min > Forget GDP. Forget cost-of-living reports. The price of a beer tells you everything you need to know about a city. How long do you need to work for one beer? That's the only cost of living metric most people actually understand. Economists overcomplicate everything. Indexes, models, baskets of goods. I've read the reports. None of them tell you what you actually want to know, which is whether you can afford to live somewhere without thinking about it. A beer answers that faster than any economist ever will. I figured this out traveling. Vietnam first. A beer at a convenience store costs 6,000 to 7,000 dong. In bulk, 5,000. At a bar, around 10,000 to 15,000. And the system is efficient. You don't order beers one by one in Vietnam. You order a crate. Twenty-four bottles. They bring it to the table. You drink what you want and they charge you for what's gone. The rest stays in the fridge for next time, or for the next group. No waste, no friction, no upselling. The price of that beer captures everything. Production, logistics, rent on the building, the staff's wages, the taxes, the margin. Especially in a bar. A beer in a bar is a compressed version of the entire local economy in a single transaction. Once you see this, you can't unsee it. Compare a few cities, ranked by software engineer salary because that's what I know and that's the lens I had while traveling. In Vietnam, a developer makes around 15 million dong a month, rent runs around 5 million, and a bar beer is 10,000. In Japan, the same role makes around 500,000 yen, rent is 100,000 to 200,000, and a bar beer is 350 to 500 yen, but you can get one out of a convenience store for 120 to 180 yen. In Toronto, the gross salary is around $90,000 a year (about $5,500 net a month after Canadian tax), rent on a one-bedroom is about $2,000 a month, and a bar beer runs $9 to $15. In France, the dev salary is around €3,500 net, rent is around €750, and a beer in a bar is €3.50 during happy hour, which in France lasts from 4 PM to midnight. | City | Monthly net | Rent (1BR) | Bar beer | Bar beers / month after rent | |---|---|---|---|---| | Paris | €3,500 | €750 | €3.50 (happy hour) | **~785** | | Tokyo (combini) | ¥500,000 | ¥150,000 | ¥150 | **~2,333** | | Tokyo (bar) | ¥500,000 | ¥150,000 | ¥425 | **~823** | | Vietnam | 15M VND | 5M VND | 10,000 VND | **~1,000** | | Toronto | CA$5,500 | CA$2,000 | CA$12 | **~291** | Different currencies. Different cultures. Doesn't matter. The ratio is what matters. How many beers does your monthly salary buy? How many beers does it buy after rent comes out? How long do you have to work to afford one drink at a bar after a long week? That's the real measure of a city. Take Toronto. Average dev salary is around $90,000 gross, which is roughly $5,500 net after Canadian tax. Rent eats $2,000 of that. So you have $3,500 left. At $12 a beer at a bar, that's about 291 beers a month of disposable income. Sounds like a lot until you also have transit, groceries, utilities, and the fact that you'd like to do other things with your life. Now take Vietnam. 15 million dong a month, minus 5 million for rent, leaves 10 million. At 10,000 dong a bar beer, that's 1,000 beers. Three times Toronto's number, and no winter coats to buy. That's the index. Not absolute purchasing power. The relationship between what you make, what you spend on shelter, and what's left over to enjoy. Now Paris. €3,500 net, minus €750 rent, leaves €2,750. At €3.50 a happy hour beer, that's 785 beers. Paris on a developer salary is, by this metric, the best deal in the developed world. Tokyo with convenience store beer prices at 120 yen comes close. Toronto is a trap. New York is worse. A beer also tells you something the salary numbers don't. It tells you about the culture. In Japan, beer is cheap in convenience stores because people drink outside. You buy a beer at the 7-Eleven, sit on a bench in a park or on the curb outside the station, and that's a normal evening. In France, beer is a social ritual. You sit at a café terrace for two hours, you nurse one beer, and the price reflects the time and the space the bar is renting to you. In Vietnam, drinking is communal. You're not paying for an experience, you're sharing a crate of cheap beer with friends on plastic chairs. The beer doesn't lie. It just shows you what the city actually values. So that's the index. The article isn't really about beer. It's about the fact that you don't need economists to tell you whether to take a job somewhere. You don't need a cost-of-living calculator. You don't need to read a single quarterly report from the IMF. You just need to know the price of a beer at the cheapest bar in town, the price of one at the corner store, the median salary for what you do, and the rent on a one-bedroom in a normal neighborhood. Run the ratios. The number is the truth. Life isn't measured in currency. It's measured in what you can consume without thinking about it. If your salary buys fewer beers than someone in another city, you're not richer. You're just paying more to exist. Here's the live version. Search a city, filter by region, switch between bar beer and store beer to see which cities are actually a deal. Find the cheapest beer in town and the rent in a normal neighbourhood, not a rough area, not a luxury tower. Median salary for whatever you do. Run the ratios yourself. Let's see where it actually pays to live. --- # Why Long Articles Are Dead URL: https://www.askgib.com/articles/why-long-articles-are-dead Published: 2026-02-02 Reading time: 5 min > I used to write 30-page articles with charts, schematics, and footnotes. Nobody read them. Here's the format I'm using instead. I had a website with two long articles on it. Each was thirty-something pages. Each took weeks to write. Each had charts, schematics, footnotes, sections, sub-sections. Each was read by, generously, almost nobody. I want to talk about why that happened, what I'm doing instead, and what this whole site is going to look like going forward. Because the failure of the long-form essay isn't a small thing. It's a sign of how reading itself has changed. When I wrote the long pieces, I thought I was being thorough. I thought I was respecting the reader by laying out every angle of an argument, citing every source, building every figure from scratch. I assumed that if I made the case in enough detail, the case would be undeniable. This was a misunderstanding of how anyone reads anything in 2026. Nobody reads thirty pages. Almost nobody reads ten pages. Most people read two screens, decide whether the writer is worth their time, and either move on or skim the rest. The detailed argument I spent weeks building gets evaluated based on the first 400 words. If the first 400 words didn't grab the reader, the next 14,000 words were irrelevant. The work of the next 14,000 words was wasted, not because the work was bad, but because the format was wrong. This isn't really about TikTok. It isn't really about social media destroying attention spans. Those are easy explanations and they're mostly excuses. What changed is that information has become abundant. In 2005, if you wanted to know something serious, you read a long article from a serious source, because that was where the depth lived. In 2026, depth is available everywhere. You can find a 90-minute YouTube video on virtually any topic. You can ask an AI to explain anything at any level. You can find an academic paper, a podcast, a Reddit thread, a Wikipedia article, all on the same topic, in the same minute. The scarce thing isn't depth. The scarce thing is the reader's time. When time was abundant and depth was scarce, the long article was the right format. When depth is abundant and time is scarce, the long article is a bad bet. The reader will get the depth somewhere else. What you can offer them is something different: a clear point, made fast, in your voice, with enough specificity to be useful and enough opinion to be memorable. The thing that actually pushed me to change my mind was reading a piece by Hakan, who is the goat at this. His article was short, sharp, and said in 1,500 words what I had been trying to say in fifteen thousand. I read it twice, agreed with him, and started rewriting everything from scratch. A piece should take **five minutes to read** at a normal pace. That's roughly **1,200 to 1,500 words** of body text. **Ten minutes is the hard ceiling.** If an article would run longer than that, it's actually two articles, or it needs cutting. It should have **a hook in the first two sentences**. Not an introduction. Not background. A specific image, a number, or a moment that pulls the reader into the argument. It should have **a clear thesis** stated within the first 200 words. The reader should know what I'm arguing before they decide whether to continue. It should be **in my voice**, the way I would actually talk about the topic over a beer with a friend who asked me what I thought. Short sentences. Real opinions. Occasional sharp lines. No hedging. No "to be fair." No "some would argue." If I think something, I say it. It should have **at most three figures**, and the figures should hit. Not decorative. Not just for visual variety. Three concrete things that the reader will remember after they close the tab. It should **end on a punch**. The closer matters more than the introduction. The last line is what people quote. And it should be **fun to write**. Not because writing should always be fun, but because if I'm not enjoying writing it, the reader isn't going to enjoy reading it. There's a tone of obligation in long-form journalism that comes through on the page. I want to write things that feel like I wanted to say them, not like I had to. | | Old format | New format | |---|---|---| | Length | 5,000–15,000 words | 1,200–1,500 words | | Read time | 30–60 min | 5 min | | Figures | 5–15 | Max 3 | | Voice | Academic / professional | Personal / direct | | Opening | Introduction | Hook | | Closing | Conclusion | Punch line | | Footnotes in body | Many | None | | Time to write | Weeks | Hours | | Readers | Few | Many | I can't make a fully rigorous case for any complex argument in 1,500 words. The Beer Index piece is a sketch. The Diamonds piece skips most of the supply chain detail. The Flying piece compresses a 200-page cost model into a paragraph. The Consultants piece doesn't enumerate every consulting scandal. Anyone who wants the full depth has to do their own follow-up research. This bothers me less than it used to. The reader who actually wants the deep version will Google it, ask an AI, or read the appendix. Most readers want to know **whether the argument is interesting enough to engage with**. That decision happens in five minutes or not at all. I'm also giving up the ability to defend the argument against every possible objection. The long-form essay can anticipate twelve objections and dismantle them all. The short essay just makes the case and trusts the reader. Some readers will object. Some objections will be valid. That's fine. The article isn't trying to be a court filing. It's trying to be a starting point for thinking. A five-minute piece in clear voice will reach ten times more readers than a thirty-minute essay on the same topic. Maybe a hundred times. The math of distribution massively favors the short, sharp version. Even if the long version is "better" in some abstract sense, the short version reaches more people, gets shared more, and influences more thinking. The second thing is that it can ship. The thirty-page essays took me weeks each. The five-minute version takes a day. The articles you're reading on this site over the next few months, there are about twenty of them queued up, would never have been written in the long format. The throughput is fundamentally different. The third thing, and this is the part I didn't expect, is that the short format forces clarity. When you have 1,500 words, you can't hide behind verbosity. Every paragraph has to earn its place. Every sentence has to be doing work. The discipline of the format makes the writing better, not worse. I should be honest about something. I was writing some of my earlier long articles with AI assistance. Not because I needed help thinking, but because I wanted to expand my ideas into polished long-form. The AI was good at it. The output was professional. It read like serious journalism. It was also boring. The earlier articles read like they were written by a McKinsey associate, not by me. The voice was flattened. The personality was sanded off. The specific moments that make me me, the trip to Vietnam where I figured out the beer index, the diamond company I almost bought, the parking garage with the broken cameras, got buried under generic prose. The new format flips this. I write the raw take, in my voice, the way I'd say it out loud. I let AI help with research, fact-checking, and polish, but the voice stays mine. The result is articles that sound like a specific human had a specific opinion, instead of articles that sound like the consensus position of every smart person in a particular field. This matters. The internet doesn't need more consensus. It has enough consensus. What it needs more of is people willing to say a thing they actually believe, in their actual voice, knowing that some readers will disagree. If you've been reading this site, what you're getting from now on is the short version. Five minutes. Three figures. A clear point. My voice. If you want the deep version, the data is in the footnotes, or you can ask. The math under the flying article is a 200-page bottom-up cost model. The diamond article rests on six months of due diligence on a French diamond company. The consultants piece is built on real DOJ settlements and McKinsey's own published research. None of that's gone. It's just not in the article. The article is the doorway. What's behind the door is up to you to chase. The age of the thirty-page essay is over. Not because depth is dead, but because depth lives elsewhere now. What I can offer that nobody else can is a specific opinion in a specific voice on a specific topic, fast. That's what this site is going to be. Five minutes at a time. As long as it takes. --- # Diamonds Aren't Timeless. They're Worthless. URL: https://www.askgib.com/articles/diamonds-timeless-but-worthless Published: 2026-01-26 Reading time: 4 min > A 1-carat diamond costs €119 on Alibaba. Your jeweler charges €900 for the same stone. Here's why. A 1-carat diamond on Alibaba costs €119. Same stone, D color, VVS2, excellent cut, at a normal jeweler? €700 to €900. That's not a markup. That's a tax on people who don't know. De Beers made up "A Diamond Is Forever" in 1947. They controlled 90% of global supply, stockpiled the rest, and told everyone diamonds were rare. They weren't. They never were. Diamonds are just carbon, and carbon is everywhere. The scarcity was manufactured. The romance was manufactured. The price was manufactured. And now the whole thing is collapsing because we figured out how to grow them in a lab. Same atoms. Same structure. Same hardness. Same sparkle. The only "fake" thing about a lab-grown diamond is the price. There are two ways to grow them: **HPHT**, squeeze carbon at insane pressure and heat until it crystallizes. Done in China. Cheap, fast, makes small stones. **CVD**, put a diamond seed in a vacuum chamber, blast it with methane and microwaves, and let it grow layer by layer like frost on a window. Done in India. Makes bigger, cleaner stones. That's it. That's the whole magic. No mantle, no billion years, no volcano. Just energy and time. The minimum order on Alibaba for a 1-carat D/VVS2 lab diamond is **one stone**. Not a thousand. One. No business license. No reseller permit. Just a credit card and an address. Here's what you actually pay, delivered to your door, April 2026: - 1 carat, **€119** - 2 carat, **€238** - 3 carat, **€390** The price scales linearly with size. Because the cost is just energy and time. A 3-carat stone takes 3x as much electricity as a 1-carat. That's it. Now look at what a regular jeweler charges for the same stone: - 1 carat, €700 to €900 *(7x markup)* - 3 carat, €3,000 to €5,000 *(8 to 13x markup)* What does that markup pay for? A GIA certificate costs €30 to €50. Shipping and insurance, maybe €20. Everything past €160 is margin. **The markup pays for ignorance.** That's the whole game. In 2000, the markup made sense, you needed someone to source the stone, explain the 4Cs, hold it up to the light. In 2026, you have the internet. The information gap is gone. The store is selling you something you can buy yourself for one-seventh the price. Pink diamonds were supposed to be impossible to grow in a lab. The industry said the color came from "plastic deformation during volcanic eruption", which is a fancy way of saying *we don't know, but it's rare, so it's expensive.* A natural Fancy Vivid Pink, 1 carat, sells for over **$1 million**. A lab-grown one, same color, same carat, same certificate? **$8,000 to $20,000.** Same stone, 98% off. Blue is even worse. Natural Fancy Intense Blue: $200,000 to $1,000,000 per carat. Lab-grown blue: $2,000 to $8,500. That's a 99% discount. Because here's the truth about lab-grown color: it's a recipe. Add nitrogen → yellow. Add boron → blue. Engineer nitrogen vacancies and anneal at high temp → pink. Irradiate after growth → green. The CVD reactor doesn't know what color it's making. The energy cost is the same. The only thing that changes is the gas mix. So when a jeweler charges you a million dollars for a pink diamond because it's "rare," remember: a machine can make the same stone next week for two grand. | Country | Energy cost per carat | Total cost per carat | |---|---|---| | China | $12–25 | $37–75 | | India | $20–40 | $48–95 | | USA | $30–60 | $90–180 | | France | $50–90 | $140–270 | | Ethiopia (future hub) | $4 | $30–50 | Now look at the prices on Alibaba again. €119 for 1 carat = ~$130. That's already at the production-cost floor for China. This is the same trajectory as solar panels. Same as LED bulbs. Same as flat-screen TVs. The cost only goes down. The technology gets better. The factories scale up. Diamonds aren't a luxury good anymore. They're a manufacturing output. And manufacturing outputs converge to cost-plus-margin pricing, eventually. By 2030, a 1-carat lab diamond will be $200 to $400 retail. By 2035, it'll be $50 to $150. And natural diamonds? They'll still exist. But not as an investment. Not as a "store of value." That part is done. A diamond loses 60-80% of its retail value the moment you walk out of the store, and the value keeps dropping every year as lab prices fall. It's the opposite of gold. Diamonds are not forever. They're depreciating assets with good PR. A few years ago, I got pulled into a group of investors looking at buying a French lab-grown diamond company called **Diam Concept**. The story is wild. Diam Concept was founded by Alix Gicquel, a professor who built France's first plasma-for-diamond research team back in 1989. Thirty years of research. Custom microwave plasma reactors she designed herself. She could grow whites, yellows, pinks, blues, the full spectrum. She had eight patents and 170 published papers. The science was world-class. In 2020, the company raised €5.2 million from Chanel, the Wertheimer family (Chanel's owners), and a couple of French VC funds. Luxury money. The kind of backing that's supposed to guarantee success. By 2022 the company was losing €700,000 a year. By May 2024 it was in receivership. We did the due diligence and the math was simple and brutal: - French industrial electricity: €0.187/kWh → **$200+ per carat in energy alone** - French skilled labor: €30K–80K per technician per year - Chinese competitor on Alibaba: 1 carat D/VVS2 for **$130 delivered** You can't compete with that. Not in France. Not at French electricity prices. Not ever. In the end, **nobody bought the company.** People bought the leftover stones, the uncut roughs, the polished pieces sitting in inventory. For pennies on the dollar. True market value. The technology? The patents? The reactors that took 30 years of research to design? Worth nothing in the jewelry market. The only viable path forward for European CVD is industrial applications, semiconductor substrates, quantum sensors, optical windows. Places where diamond's *purity* matters and the buyer pays for physics, not for sparkle. Alix Gicquel had the best diamond technology in Europe. She lost to the electricity bill. That's when I knew: this industry is over. Not slowly. Not eventually. **It's already over**, most people just don't know it yet. If you're getting engaged, get a lab-grown stone. Save 80%. The diamond is identical. The only thing you're losing is the markup. And if you already have a natural diamond, don't sell it. Just keep it for what it is: a beautiful rock with a great story. Just don't pretend it's an investment. Diamonds aren't timeless. They're just expensive until they aren't. --- # The Land Was Amazigh First URL: https://www.askgib.com/articles/the-land-was-amazigh-first Published: 2026-01-19 Reading time: 4 min > I am Kabyle. The border called Algeria is recent; the people under it are not. A look at the oldest layer of North Africa, and the diaspora that quietly folded into France. I am Kabyle. That word means more to me than the line on the map that says Algeria, because the line is recent and the people under it are not. Algeria is a border. France drew most of it. Inside that border are peoples who never asked to be one thing: the mountain Kabyles, the Chaoui of the Aurès, the Mozabites, the Tuareg of the deep south. Different climates, different histories, different things held sacred. The map flattened all of it into a single name, and the name is the youngest thing in the room. Start with what was here first. The Amazigh, the word means FREE PEOPLE. Their script, Tifinagh, was being carved into stone in the sixth century before Christ. Numidia was a kingdom with its own kings, Massinissa and Jugurtha, while Rome was still deciding what it wanted to be. None of this is borrowed. It is the floor everything else got built on. Then everyone came. Phoenicians traded on the coast. Rome ruled and left its roads and cities. North Africa became one of the great centres of early Christianity, and it held Jewish communities for centuries, whole Berber tribes that practiced Judaism. When the Arab armies arrived in the seventh century, the person who met them was a Berber queen, Kahina, out of the Aurès. She was Jewish. Her people were pagan and Christian and Jewish all at once, and she fought the conquest to the end. Then the Arabs stayed, and later the Ottomans, and then France. Each one left something behind. Here is the part people forget. The Amazigh were never erased by any of them. They chose. They kept what fit and let the rest wash through. You see it most clearly in how they hold the sacred. The old Berbers venerated their dead as something close to gods. That same instinct became the cult of Christian martyrs, and then became the marabouts and the saint shrines that people in Kabylie still walk to today. The names changed with every religion. The practice is older than all of them. Go to certain trees, certain mountain passes, certain tombs, and you are standing somewhere that was holy before Islam, before Christianity, before any of it had a name. That is the whole point. The religions passed through. The people stayed, and so did the oldest layer underneath them. You can stand in one place and see all of it at once. Near Tlemcen, in the west, is the tomb of Rabbi Ephraim Encaoua, a Jewish sage who fled the persecutions of Spain in the 1390s and, the story goes, arrived riding a lion. For centuries his grave was the great Jewish pilgrimage of North Africa. Every spring, around Lag BaOmer, the pilgrims came in their tens of thousands and nearly doubled the town, and not only Jews. Muslims walked to the same tomb to ask the same favours, Berbers among them, everyone borrowing everyone's saints. That was the country once. A land that made room. The last great pilgrimage was in 1955, and then the war came and the Jews left and the gate quietly closed. The south is not the north. Desert people kept different influences than mountain people, and that is not a flaw, it is the proof. A single, fixed identity could never have survived this many empires. A layered one did, easily. Which makes the last stretch hard to watch. At independence, people thought Algeria would become one of the great nations. A country that kept the best of what France left and built it on a Berber foundation. It had the land, the oil, the people, the history. Instead the room got smaller. A people who had spent three thousand years choosing were told there was one way to be, one language to prize, one faith to show in public. You can measure what that cost. The minimum wage is 24,000 dinars a month, raised at the start of 2026 after sitting frozen at 20,000 since 2020. At the real street rate for the euro, now near 290 and still climbing, that is about 83 euros. The official rate would call it closer to 160, but no one can buy euros at that price, and the dinar falls faster than the wage rises. Eighty-some euros a month, on top of some of the largest gas reserves on earth. So people left, and Kabyles left first and in the greatest numbers. The early waves of migration to France were heavily Kabyle, which is why the accent still carries through certain French towns. Around five million Kabyles live in Algeria today. How many live abroad, no one can really say, and that is the nature of it. Families were large, often a dozen children to a household, so a single generation could turn a handful of siblings into dozens of grandchildren, and every one of them born abroad is born French, French in passport and in how they move through the world. Multiply that across a century of migration and, counting the second and third generations, perhaps ten-fifteen million people of Kabyle descent now live abroad. The diaspora is large and almost invisible at once, because we assimilate so completely that by the second generation we have folded into the country around us, climbing high in school and work and then going quiet about where we came from. That is the most Kabyle thing about it. And it was never that the people who left lacked anything. They are quick, they learn fast, they adapt to any country that will have them, and they prove it the moment they land somewhere that lets them work. They left because the place that should have been theirs offered no future, only a smaller and smaller room. A land that once made room for everyone, that took in Phoenician and Roman and Jew and Arab, now cannot hold its own. That is the part that should sting. Not that the people lacked the capacity, but that they were stuck in a situation far beyond their power to fix, and so they did the rational thing and carried the oldest layer somewhere else. But the floor is still there. You cannot un-carve Tifinagh. You cannot un-build Numidia. You cannot un-visit a shrine that was holy before the prophets. The oldest layer is the most durable one, and it is still the first thing a Kabyle recognizes in himself, before any flag and before any creed. I did not go looking for this history to feel proud. I went looking because the official story is too small. The real one is older, stranger, and freer, and it is still right under our feet. --- # The Living Soil: Why What's Beneath Your Feet Matters More Than You Think URL: https://www.askgib.com/articles/the-living-soil Published: 2026-01-12 Reading time: 2 min > Soil isn't just dirt, it's a complex ecosystem teeming with billions of microorganisms. Understanding the biology beneath our feet could be the key to more resilient, productive agriculture. Most people think of soil as inert ground, the stuff you walk on, that plants grow in, that gets muddy when it rains. The reality is far more interesting, and far more important. A single teaspoon of healthy agricultural soil contains more microorganisms than there are people on Earth. Bacteria, fungi, protozoa, nematodes, and arthropods form an intricate web of interactions that shapes plant health, water retention, carbon storage, and nutrient cycling. When we talk about "healthy soil," we're really talking about the health of this invisible ecosystem. The concept of the **soil food web** was popularized by soil ecologist Elaine Ingham in the 1980s and 90s. The basic idea: every organism in the soil is eating something, and being eaten by something else. Bacteria decompose organic matter; protozoa eat bacteria and release nitrogen in plant-available forms; fungi form symbiotic relationships with plant roots to extend their effective reach for water and phosphorus. This web is self-regulating in healthy soils. Disturbances, tillage, synthetic fertilizers, pesticides, can collapse parts of it, leading to a cascade of consequences: reduced water infiltration, increased erosion, higher fertilizer requirements (because the biological nutrient cycle is disrupted), and less resilient crops. One of the most remarkable relationships in soil biology is the **mycorrhizal association** between fungi and plant roots. Over 90% of land plants form these partnerships. The fungus extends filaments called hyphae far beyond the root zone, effectively multiplying the plant's surface area for nutrient and water uptake by orders of magnitude. In return, the plant feeds the fungus carbon, up to 30% of photosynthetically fixed carbon in some species flows directly to mycorrhizal partners. It's an expensive partnership, but the trade-off is worth it: plants with intact mycorrhizal networks are measurably more drought-tolerant, disease-resistant, and phosphorus-efficient. The catch? Synthetic phosphorus fertilizers suppress mycorrhizal colonization. When phosphorus is abundant, the plant has no incentive to maintain the energetically expensive partnership. This is one reason why conventional high-input agriculture creates a dependency cycle: you disrupt the biology, so you need more inputs, which further disrupts the biology. For the last century, agricultural soil was valued almost entirely by its chemical composition, nitrogen, phosphorus, potassium, and pH. Soil organic matter (SOM) was noted, but rarely the primary focus. That's changing. Soil organic carbon is now understood to be central to almost every aspect of soil function. Higher SOM means: - Better water-holding capacity (each 1% increase in SOM allows soil to hold roughly 20,000 gallons more water per acre) - More microbial activity and diversity - Improved aggregation (soil structure) - Long-term fertility Regenerative agriculture practices, cover cropping, reduced tillage, compost application, diverse rotations, are fundamentally about building organic matter. It's a slower, less immediately measurable approach than applying a fertilizer bag, but the compounding returns over years and decades are substantial. 1. **Minimize soil disturbance.** Every time you till, you shatter fungal networks and oxidize organic matter. If you can no-till or reduce tillage frequency, the biology will thank you. 2. **Keep the soil covered.** Bare soil is biologically inert and erosion-prone. Cover crops, mulch, or even weeds are better than nothing. 3. **Feed the biology, not just the plant.** Compost, manures, and green mulches feed the microbial community. Chemical fertilizers feed the plant but often bypass or suppress the biology. 4. **Be patient.** Rebuilding soil biology after intensive management takes years, not weeks. But the trajectory matters more than the starting point. The soil is doing enormous work for us, mostly invisible and mostly for free. Understanding that work is the first step toward managing land in a way that doesn't spend down that capital faster than it accumulates.