--- title: "Paperwork with a stethoscope" author: Dr Arman Ouveysi date: 2026-07-07 room: Lab tags: medical-ai, medicine, futurism canonical: https://drouveysi.com.au/essays/paperwork-with-a-stethoscope/ summary: "In one January week, the first state-approved AI prescription-renewal pilot opened and the big labs connected their models to medical records. What actually got automated, what the automation reveals about the job, and what is left of the doctor when the routine is gone." --- # Paperwork with a stethoscope *Everything I say here about Doctronic's system comes from the public record, and the opinions are mine, not my employer's.* In the space of one week in January 2026, a machine was authorised into the prescribing loop and the two biggest AI labs shipped ways to connect their models to people's medical records. I watched it from an unusual seat. Two days a week I'm a GP in Melbourne. The rest of the week I'm an engineer at Doctronic, the company whose system had just become the first AI a US state has authorised to take part in prescription renewals, under Utah's regulatory sandbox. For a year my two jobs had been circling each other politely. That week they stopped circling. The same week, OpenAI shipped ChatGPT Health, which lets people connect their actual records and results, and Anthropic followed within days with a version for Claude. So the general-purpose models are no longer answering health questions in the abstract. They can see your data. Prescribing made the headlines, and earned them. The records access arrived with far less noise, and it deserves more attention than it got; I'll come back to why. The commentary sorted itself into the two default positions within hours. One camp announced the beginning of the end of doctors. The other performed the ritual reassurance that AI is just a tool that will free clinicians to focus on patients, a sentence that has been copy-pasted into every health-technology press release since the fax machine. Both camps missed what the week actually did. It didn't replace anything. It subtracted something, and the subtraction is far more interesting than a replacement would have been. Here is the unglamorous truth about prescribing. Most of it is renewal. Depending on how and where you count, somewhere between six and eight scripts in ten are not decisions but continuations: a person on the same statin for five years, stable, doing fine, back in the waiting room because the repeats ran out. There is no clinical question in that appointment. There is a queue, an item number, and a signature. It is not medicine. **It is paperwork with a stethoscope draped around its neck.** And it is not harmless paperwork, because it consumes the scarcest resource in the system, which is clinical time. Renewals fill appointment books for weeks. People who can't get in stop taking medication that was keeping them well, not because anyone decided they should, but because the queue decided for them. Their numbers drift, and eventually some of them arrive in an emergency department with a problem that began life as a scheduling failure. We have let an administrative bottleneck impersonate clinical care for so long that the costume fooled us too. So when a machine takes that work, I don't feel replaced. I feel the specific relief of putting down a box I had been told was mine to carry. Nothing in the years of training was about repeats. The years were about sitting with someone in the middle of a breakdown, catching the thing they're not saying, holding complexity without rushing to close it. Renewals were never the job. They were the tax on the job. I should make the honest case against my own relief, because there is one. The renewal appointment was also medicine's accidental screening program. The script was the pretext; the visit was the surveillance. The blood pressure got checked because the person happened to be in the room. The "while I'm here, doctor", delivered at the door handle, was now and then the most important sentence of the day. Subtract the routine visit and you subtract that ambient safety net with it. This is a real cost and it deserves better than a wave of the hand. But look at what kind of safety net it was: rationed by friction, distributed by luck, and thinnest exactly where the need was greatest, because the people most harmed by the bottleneck were the ones who never got through it. If opportunistic checking earns its place, and it does, the answer is to design it deliberately rather than smuggle it inside a queue. A safety net you can only reach by waiting three weeks for a signature is not a design. It is an accident with good PR. Strip the routine away, though, and you're left holding a sharper question than the replacement debate ever asks. Not "will AI take the doctor's job", which assumes the job is one thing. Rather: when the automatable majority of the job is gone, what remains, and is what remains a profession? I think there are two answers, and they point in different directions. The first answer is the one every essay in this genre reaches for, so let me state it properly before complicating it. The word doctor comes from the Latin docere, to teach, and that etymology has always felt closer to my actual working day than anything on the billing schedule. A large share of what I do is translation: taking something frightening and unfamiliar that is happening inside a person's body and turning it into something they can hold. And a share of it is weight-bearing. Early in training, somebody assured me that delivering bad news gets easier with experience. It doesn't. It gets harder, because now I know the people. I know their stories, their families, what they were quietly hoping the scan would say. The weight doesn't lighten, and I've stopped wanting it to, because the connection that makes those conversations heavy is the same connection that makes the work mean anything. You don't get one without the other. ![Doctor's work unbundled into document-shaped and judgement-shaped tasks](https://drouveysi.com.au/assets/articles/lab/paperwork-with-a-stethoscope/fig-01.svg) That is the empathy answer, and it is true, and on its own it is not enough. Notice what it concedes. If the human contribution reduces to warmth and translation, then the intellectual core of medicine, the actual figuring out of what is wrong and what to do about it, has been quietly handed to the machines, and the doctor has been re-badged as a kind of empathic interface. **I did not train for a decade to become a warm login screen.** The empathy answer is where most essays stop. It's where this one starts getting interesting, because the second answer runs the other way. Think about the patients medicine serves worst. People with fibromyalgia, chronic fatigue syndrome, irritable bowel, functional neurological disorder. People with complex, overlapping neurodevelopmental presentations that refuse to sit inside one tidy label. A large part of my clinical week is spent with exactly these patients, the ones other doors have closed on, and the longer I do it the more certain I become of one thing: they are not failing to get better because they are difficult. They are failing to get better because our instruments are not good enough, and I am counting our concepts among the instruments. **A diagnosis is a compression algorithm.** It takes the high-dimensional mess of one person's biology and history and returns a short label that is supposed to predict course and treatment. When the underlying structure is simple, the compression is magnificent: strep throat throws away a million irrelevant details and keeps the single decisive one. But for a large class of patients, the compression is lossy in precisely the places you can least afford it. Fibromyalgia does not name a mechanism. It names a cluster, a statistical neighbourhood, our best current attempt to draw a circle around people whose suffering co-occurs. Useful, sometimes. An answer, no. Women have been living inside this failure for generations: told the pain was anxiety, the fatigue was stress, the symptoms didn't add up to anything real. The problem was never them. The problem was a discipline that lacked the resolution to see what was there, and that preferred to doubt the patient rather than doubt its own categories. Autism teaches the same lesson from a different angle. We use one word for a person running a trillion-dollar company and a person who cannot speak, and we call the distance between them a spectrum. A spectrum that wide is not a description. It is a placeholder, an IOU written against future understanding. Here is the part I find genuinely hopeful, said with all the caution it deserves. The reason medicine compresses so aggressively is not laziness. It is bandwidth. A clinician can hold perhaps half a dozen variables in working memory while reasoning under time pressure. The patients I'm describing present with hundreds: a decade of results, medications tried and abandoned, life events, family patterns, environments, the lot. They were never too difficult. They were too high-dimensional for the instrument examining them, and the instrument was a human brain with fifteen minutes. That specific constraint, the one that has quietly shaped every diagnostic category we have, is the constraint that just moved. A model that can genuinely attend to a whole record, and as of that January week the whole record is exactly what these systems are beginning to see, is the first plausible instrument for finding structure inside the labels rather than around them. This is why I think the records access was the bigger news. The prescribing decision automated medicine's most routine act. The records access points, eventually, at its hardest one. Now the caution. Nothing about current systems guarantees any of this. Today's models compress toward the textbook, because the textbook is what they were trained on, and for exactly these patients the textbook is the problem. The nightmare version is easy to build and is probably being built somewhere right now: the old dismissal, automated, delivered at scale with infinite patience and perfect grammar. "Your results are normal, have you considered stress", a hundred million times, in a reassuring voice. Whether we get the telescope or the rubber stamp is not destiny. It is a design and evaluation question, which is to say it is the exact territory the rest of this room is about, and the reason I spend my week on evals rather than adjectives. Play it forward a decade and the shape of the profession changes in a way I find easier to welcome than to fear. Genomic analysis in the consultation rather than the referral letter. Imaging read in the room. Results on the spot instead of next Tuesday. Procedures done by systems steadier than any hand. In that world the machines own precision, and precision was never the human comparative advantage anyway; ask anyone who has read my handwriting. What the human owns is integration: deciding what all of it means for this particular person, with their particular fears and finances and family, and then standing next to them while it means it. There is an inversion hiding in that future which I haven't seen anyone say out loud. For a century, prestige in medicine has tracked narrowness. The generalist sat at the bottom of the status gradient and the super-specialist at the top, because depth was scarce and human heads were the only place depth could live. If depth becomes something you can summon on demand, the gradient flips. The scarce skill stops being knowing one thing completely and becomes moving across everything competently: from broad unknowns to specific answers, from vague plans to targeted ones, with an immense array of instruments and the judgement to know which one this moment needs. The last doctor standing looks a lot less like a subspecialist and a lot more like **a general practitioner with superpowers**. I concede that this conclusion suits me suspiciously well. I have examined it for motivated reasoning and I'm keeping it anyway. Something shifted in that January week. I felt it from both chairs. And the feeling, when I checked, wasn't fear. It was the thing that got me into both of these jobs in the first place: curiosity. The routine is leaving the profession, faster than most of the profession believes. I want to find out what kind of doctor is left when everything I do out of necessity has fallen away, and all that remains is the work I do because it is worth doing.