Every tool that has ever been sold to a clinician has been sold the same way: this will give you your time back.
So it is worth slowing down when the receipts finally arrive. Over 13 months and more than ten thousand emergency department encounters, Stanford measured what an ambient AI scribe actually does to a physician's clock. Not perceived. Not surveyed. Audit logs.
The answer is genuinely good, and genuinely smaller than the brochure. About 72 seconds per patient. Roughly 24 minutes across a shift. Real minutes, honestly earned, and nothing like the transformation the category has been promising since 2023.
This Sunday we are looking at the gap between those two numbers, because that gap is where most of healthcare's disappointment with technology lives. A tool can work exactly as designed and still not fix the thing you hoped it would fix.
The study is worth your attention precisely because it is unflattering to everyone, vendors and skeptics alike.
Let's get into it.
— Troy, Ray, and Ibrahim
Your AI scribe saved 72 seconds. Where did the evening go?
Researchers at Stanford published a retrospective cohort study in JMIR AI covering a staged ambient AI scribe rollout in a busy tertiary academic emergency department, running from February 2025 through March 2026. The analytic cohort: 10,344 encounters handled by 100 attending physicians, with use left entirely to physician discretion, encounter by encounter.
Using mixed-effects models with physician random intercepts, the team found ambient scribe use was associated with a 72.6 second reduction in on-shift documentation time per encounter (95% CI 63.8 to 81.4, P<.001). Across 20 encounters, that is about 24 minutes back in an 8-hour shift.
Why it matters: Almost everything published on ambient documentation so far has been a vendor pilot, a satisfaction survey, or a few weeks of enthusiastic early adopters. This is a real rollout, measured against audit logs, adjusted for patient and physician factors, with negative control outcomes and a placebo permutation test to check that the effect is not just confounding. It is the closest thing the category has to honest evidence.
The catch: Read the rest of the table. After-shift documentation time went up by 9.1 seconds, and notes got 690 characters shorter. A separate 60-day pilot of 40 clinicians found something more uncomfortable still: large, significant drops in perceived documentation time and after-hours work, while objectively logged EHR minutes did not move at all (P=.15). The tool reliably changes how the work feels. Whether it changes how long the work takes is a different question, and the two studies do not agree.
Bottom line: Twenty-four minutes a shift is worth having. Just do not confuse it with getting your life back, and do not let anyone sell it to you as though it were. When the next contract comes up for renewal, ask for the audit-log data, not the satisfaction scores.
Read the full report from JMIR AI →
Where do you land on this one? Hit reply and tell us. We read every response, and the best ones shape the next issue.
From the Field
This week we are asking about the gap between how a tool feels and what it actually does. Has ambient documentation changed your day, or just changed where the charting happens? We want to hear from the attending who finally stopped taking notes home, and from the one who found the work simply moved to a different hour. Nurses, residents, PAs, and the informatics people who ran the rollout: all of it counts. We protect your anonymity and your patients' privacy, always.
The Profile: Shiv Rao, MD
Cardiologist at UPMC, co-founder and CEO of Abridge
Shiv Rao did not come to ambient documentation from software. He came to it from clinic, as a practicing cardiologist at the University of Pittsburgh Medical Center who spent his evenings and weekends finishing notes, and decided that was a problem worth building a company around.
He still sees patients. That detail matters more than it should have to, in a category where most of the people designing the tools have never carried a panel.
- Co-founded Abridge in 2018 while practicing cardiology at UPMC, building the platform around medical conversation rather than general-purpose transcription
- Won a seat on the Department of Veterans Affairs enterprise contract in September 2026, valued at up to $775.7 million across awardees
- Deployed across more than 75 VA Medical Centers during the pilot phase, spanning primary care, twelve medical and surgical specialties, and the VA's Clinical Resource Hubs serving rural veterans
- Now in use at more than 150 health systems nationwide
Quick Hits
The VA goes all in on ambient AI
Abridge won a seat on a Veterans Affairs enterprise contract worth up to $775.7 million, expanding from a pilot across 75-plus VA Medical Centers to national deployment. It is the largest single public-sector commitment to ambient clinical documentation to date, and it lands in a system still working through EHR modernization. LEARN MORE
Providence puts numbers on the productivity question
A research team at Providence St. Joseph Health published an analysis in JAMA Network Open examining ambient AI use against clinician documentation burden, productivity, and efficiency. Health systems weighing enterprise contracts now have a second large operational dataset to argue with. LEARN MORE
Perception moved. The clock did not.
In a 60-day pilot of 40 clinicians, perceived documentation time and after-hours work dropped sharply, with large effect sizes, and NASA Task Load Index scores fell across mental demand, effort, time pressure, and stress. Objectively logged EHR documentation minutes did not change significantly. Both findings are real, and they are not the same finding. LEARN MORE
The Joint Commission starts certifying AI governance
The Joint Commission launched its voluntary Responsible Use of AI in Healthcare certification in June 2026, built on guidance developed with the Coalition for Health AI. Any organization can apply, accredited or not. It is the first structured answer to a question most health systems have been improvising: who signs off on the model? LEARN MORE
Off the Clock: the evening you were promised
There is a line that has been circulating for a couple of years now: the most valuable thing AI can give a doctor is not a smarter diagnosis, it is their evening back.
It is a good line. This week complicates it in a useful way.
Because the Stanford data says the minutes are real, and the Minnesota data says the relief might be mostly in how the day feels. Those are not the same gift. One is time you can spend. The other is a lighter cognitive load, which is not nothing, and which the NASA Task Load Index picked up loudly: less mental demand, less time pressure, less stress. Clinicians in that pilot were not lying. They were describing something the audit log cannot see.
Maybe that is the honest framing. The scribe is not buying back your evening. It is lowering the cost of the hours you are already working. Anyone who has finished a shift feeling wrung out rather than merely tired knows those are different currencies.
But here is the part worth sitting with. If the relief is partly perceptual, it is also fragile. Perception adjusts. The thing that feels like a reprieve in month two becomes the baseline by month eight, and then the panel grows, or the staffing thins, or the template gets longer, and the 24 minutes quietly get spent by someone who is not you.
That has happened before, with every efficiency medicine has ever found. The productivity gain does not stay with the clinician by default. It stays with whoever is allowed to claim it.
So the question underneath the documentation debate was never really about documentation. It is about who owns the time when it comes back. That one no model is going to answer for us.
Until next Sunday.
For the people keeping medicine human.
