For a century we paid for the answer and tolerated the manner. As machines make the answer cheap, the scarce thing left is the part we always undervalued: getting a frightened human being to actually follow the plan.
We treat a clinical note as thorough only if a doctor typed every word of it. We have confused the pain of writing with the quality of the record — and we keep rewarding the wrong thing.
A resident scribbling potassium values on the back of their hand is not careless. They are out of working memory, and a ballpoint pen on skin is still a better interface than the record we built them. A case for automating the bureaucracy so the mind is free to reason.
An AI that reads a scan and an AI that summarises a history are not the same kind of machine, and they do not earn our trust the same way. One we validate against trials. The other we must be able to check, line by line, against its source.
We hire some of the most expensive minds in the economy and spend half their day on tasks that require literacy, not a medical degree. That is not a wellness problem. It is a misallocation of capital we would never tolerate anywhere else.
Show a hospital a problem and its first instinct is to add something to the room: a tablet, a sensor, a robot in the lobby. The harder, better question is what we could take away. A case against electrified clutter.
For a century, medicine rested on a quiet asymmetry: the doctor held the facts, the patient did not. That asymmetry has dissolved. A patient with a phone now reaches the same survival curves as the professor. So what is the doctor for?
We track bed occupancy, cash flow and medication stock to the decimal. The one resource we never measure is the one most likely to run out on a ward: the attention of the person at the bedside. What we decline to count, we quietly allow everyone to spend.
Hospital dashboards worship speed: door-to-doctor, length of stay, throughput. We buy technology to remove every pause between symptom and prescription. But in medicine the pause is not a defect in the line. It is the work.
Anxious young doctors spend their weekends learning to code, sure it is how they survive the AI wave. A rational reaction to a failing curriculum, and the wrong problem. The skill that matters is not building the tool but knowing when it lies to you.
Every patient in a modern hospital has a double: a digital profile assembled to fit the reimbursement rules. Our safety systems increasingly consult that double rather than the person in the bed. When the codes are wrong, the consequence is no longer just an audit risk.
We built clinical software around a patient who does not exist: one clear complaint, one tidy pathway. Real patients arrive in the plural. When the form will not bend, the clinician bends the truth instead — and clicks the least wrong box.
When old software failed, it failed loudly — garbled text you could not miss. Generative AI fails in flawless prose. The error no longer announces itself, and a perfectly written note may be the most dangerous thing in the record.
The patient moves from the GP to the specialist to the ward and back. Their record does not. Every threshold they cross, they leave a part of their history behind — and we treat the resulting blindness as a plumbing problem rather than a safety one.
For a century we selected doctors for what they could memorise, because the physician was the storage medium. That problem is solved. The skill that remains scarce is the one we barely test for: making sense of a patient when the data disagrees with itself.
Healthcare technology is sold on minutes saved. But a minute spent logging in at eight in the morning is not the same as a minute spent deciding on an end-of-life plan at four in the afternoon. The scarce resource is not time. It is judgement.
Watch a doctor on rounds and the frustration is rarely about the medicine. It is about the screen, which interrupts clinical thought to confirm a billing code. We did not digitise the chart. We digitised the invoice and taped the patient's history to the back.