During my first few months as a newly minted primary care physician, I spent many weeknight evenings negotiating with myself. The negotiation was always the same. If I close three more notes, I’ve done enough to go home for dinner. Enough to have earned it.
I knew, while making the deal, that “enough” was a fiction, because after dinner I’d open the laptop again and work until the work was done, or I was. Dinner was a brief reprieve because my second shift was waiting behind it.
Most physicians and APPs I know run some version of this negotiation. Our struggles with charting and administrative tasks are more akin to a debt, carried from the exam room to our workstation to the parking lot to the kitchen table, accruing interest the whole way.
For a few years now, the promise has been that artificial intelligence would retire that debt. Ambient AI scribes listen to the visit and draft the note, and by most estimates roughly a third of clinicians now have access to one, with the majority expected to by the end of this year. The pitch is clean: Talk to your patient, walk out with a finished note, get your evenings back.
Real physician voices, twice a week
Free, and one click to unsubscribe.
Then, in April, a study published in JAMA challenged that assertion. It tracked more than 8,000 ambulatory clinicians across five academic medical centers from 2023 to 2025, about 1,800 who adopted scribes against roughly 6,800 who didn’t. The result, reported by STAT that same week, was that adopters saved about 16 minutes of documentation time and 13 minutes of total EHR time per eight hours of patient care. Real, measured, modest.
Sixteen minutes.
That number disappointed a lot of us, but it shouldn’t have surprised us. The same JAMA study found no significant reduction in after-hours EHR time, the exact hours we were promised back.
Among the clinicians who adopted a scribe, only about a third used it in half or more of their visits, the threshold at which the benefit is seen. Those heavy users got roughly twice the reduction in EHR time and three times the reduction in documentation time. Everyone else picked the tool up and put it back down. Said differently, the technology works when it’s used well and used consistently, and most physicians aren’t doing either.
What’s missing are the workflow designs, habit changes, and protocolized ways of actually closing work in the room. This gap has a name, the adoption layer, and AI scribes don’t ship with those.
A 2026 analysis in JMIR Medical Informatics noted that reported time savings vary considerably between organizations, which is what you’d expect if the tool were doing less of the work than the implementation around it. And the study’s own authors concluded that clinicians may need guidance to benefit. An unusually candid admission that the tool alone isn’t the answer.
It gets less reassuring. Vendor positioning has quietly shifted toward revenue-cycle and coding value, more patients per session, cleaner billing. Worthy goals for a balance sheet. But they are not the same goal as getting a physician to dinner without a laptop, and we shouldn’t assume the first one inevitably drags the second along behind it.
Here’s what I’ve watched happen when the adoption layer is supplied. Across more than 130 physicians and APPs over eight cohorts, structured workflow and behavior-change training cut after-hours charting roughly in half, from about 13 hours a week to under 7. At the same time, high-burnout prevalence fell from most of the group to about half of it, and professional fulfillment roughly doubled.
No new software required. Just a different way of thinking about the work combined with strategies for sequencing and closing the work. The technology and the training aren’t rivals; they’re complementary. The training is what lets the technology deliver what it promised.
So, two takeaways, because two different people are reading this.
If you’re a physician or APP: The tool can amplify a workflow, but it doesn’t install one. A scribe pointed at the same habit that generates your debt will hand you a slightly cheaper version of the same debt. The lever you own tonight is how you sequence and close a visit, whether the note, the orders, the messages, and the billing leave the room when the patient does. That’s learnable, with or without AI.
If you lead a health system: You are buying software licenses. You are probably not buying the adoption layer, and the evidence suggests this layer is where the evenings live. Fund the training and the workflow redesign the way you fund the seats. And measure the thing that matters, after-hours time, not vendor logins. Sixteen minutes is what you get when you buy the tool and skip the change. It’s a real sixteen minutes. It’s just not the evening back.
I am grateful to say that thanks to implementing the adoption layer I no longer negotiate with myself on weeknights. My evenings are my own, with or without an AI scribe.
This essay is cited in the KevinMD record on the electronic health record.
Junaid Niazi is an internal medicine physician.
