The End of Generalist Doctors

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.

This isn't a guess. The studies are out.

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, across pooled data from over 70,000 test images, reach 91% sensitivity and 88% AUROC for melanoma detection.

Source: BMC Primary Care 2025 meta-analysis; 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 European Journal of Medical Research 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.

Let me be precise about what a general practitioner spends their day doing.

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.

I get the instinct. But look at what the current system actually delivers.

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 average wait to see a specialist after a GP referral is 27.7 weeks as of 2024, 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.

Like with lawyers, generalist doctors aren't going to disappear. The role just contracts dramatically.

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.

There's a real risk in this transition that's worth naming.

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.