Who keeps the minutes?

A friend of mine ran her clinic on back-to-back fifteen-minute slots for years. Last spring her practice rolled out an ambient AI scribe. The notes finished themselves now, the pager went quiet after four, and for about six weeks she left the building while the sky was still light. Then the schedule quietly grew by two patients a day. Nobody announced it. The slots just got shorter, the way boiler pressure creeps up until somebody checks the gauge.

That trade, time invented and time immediately invoiced away, is how the business model already runs.

The productivity dividend in medicine is real, and the institution confiscates it before the clinician ever banks it.

The evidence has stopped being anecdotal. The AMA's CEO Dr John Whyte put it on the record in the On/Offcall newsletter: AI efficiency gains get redirected into higher patient volumes rather than less burnout. A peer-reviewed paper published on 8 September, from McGill, SingHealth and Duke-NUS researchers, gave the confiscation a clinical name: "productivity co-optation". Once scribes document time savings, scheduling templates expand and throughput targets creep up, "effectively transferring the efficiency dividend from clinician well-being to organizational output."

The harvest is getting quantified too. In the Philips Future Health Index survey, 36% of US healthcare professionals said AI increased their capacity, at a median of five extra patients a week. Five more patients a week is more work for the same doctors, not more time at home.

The gift that proves it's a choice

Here's what should bother every administrator quoting the ROI slide. On a Harvard Business School case podcast about Mass General Brigham's 12,000-physician scribe programme, one physician recalled how human scribes used to work: the human scribe had to be paid for, so the doctor had to see more patients to cover the cost. Economist Susanna Gallani's warning was blunt: keep that logic with AI and the purpose of reducing burnout just goes out the window.

Cleveland Clinic ran the opposite experiment. In four months it onboarded more than 4,000 ambulatory clinicians onto an ambient scribe, framed "as a gift rather than mandate": no requirement to see more patients or take on extra work in exchange for access. Sixty percent said it raised their likelihood of staying in practice. The economics closed through better coding and billing, not through extra bodies in the schedule.

Same technology, same time saved, only the owner of the minutes changes.

Ownership follows tooling

AI does save clinicians time, though less than the marketing says: an Intermountain matched-cohort study reported no statistically significant productivity gains, and the honest surviving benefit is lower cognitive load. What actually gets contested is who owns the saved minutes, and ownership follows tooling.

A clinician using the hospital's AI has no claim on the dividend, because the hospital owns the tool. Build the tool yourself and the claim comes with it, the same way it works in every other industry: the person who owns the machine decides what the hour is worth.

On Saturday 3 October in Antwerp, Care & Code puts about 75 clinicians in a room with AI build tools for Belgium's first clinical vibe-coding day. I co-organise it, so read the next line as an invitation with skin in the game. But the premise is exactly this ledger: teaching doctors to build their own tools is how they keep what the technology frees up. The prior-auth generator a clinician writes in a morning is the one the hospital can't quietly convert into two extra clinic slots.

When your hospital deploys its next scribe, ask the only question that matters: who keeps the minutes?

If the answer is a throughput target, you know what the deployment is for. If clinicians can't answer, they don't own their own time yet. The scribe isn't the product. The answer to that question is.

💥 May this inspire you to count the minutes before someone else does.