Ambient documentation, two years in. What the physicians who kept using it actually say.

Ambient clinical documentation is now the most widely deployed generative AI application in medicine. Adoption inside large US integrated delivery networks moved from single digit penetration in mid 2024 to something in the range of a third of eligible physicians by mid 2026, with several large systems reporting majority uptake among primary care clinicians.

The vendors' case is well rehearsed. Reduced after hours documentation time, improved coding capture, measurable reductions in self reported burnout. Each of those claims is supported by peer reviewed evidence, though the effect sizes vary substantially by specialty and by implementation quality.

What is less well covered is the view from the physicians who have used the technology daily for eighteen months or more. In structured conversations with twenty two such clinicians across the US and UK, three themes recurred.

First, the time savings are real and durable. None of the physicians we spoke to would voluntarily return to writing their own notes.

Second, the character of the consultation has changed. Several physicians reported that they now speak differently in the room, more structured, more explicit about differential reasoning, because they know the note is being generated from what they say. Whether that change is net positive or negative for the patient is contested and largely unstudied.

Third, and least discussed, the technology has begun to change the training pipeline. Residents entering practice in 2026 are the first cohort to have trained partly with ambient documentation in place, and the concern raised most frequently by senior physicians was that these residents may not have developed the note writing skills their predecessors did.

None of this is a case against the technology. It is a case for coverage that treats it as the first genuinely mature clinical AI application, rather than the last novelty.