If the ambient scribe is writing the first draft of the chart, where does the resident learn to reason through a patient? Program directors are starting to ask that out loud, and nobody has a clean answer yet. For a generation of trainees, the note was the classroom.
You sat down after a visit, pulled the history together, forced yourself to prioritize, and found out what you understood and what you only half-understood. A faculty attending marked it up and sent it back. The deliverable and the training were the same piece of paper.
That piece of paper now gets generated in under a minute by software that heard the whole encounter. The chart gets done faster. The learning loop that used to ride along with it does not, unless someone deliberately rebuilds it. A recent Chicago Journal essay on what happens to the on-ramp when AI writes the first draft frames the same training problem across industries, and in medicine it shows up in a few distinct shapes, each one asking for a different response.
The Resident Who Becomes a Passive Editor
The most common pattern is the trainee who stops writing and starts approving. The scribe produces a plausible note, the resident skims it for anything obviously wrong, clicks sign, and moves on. The documentation gets done. The reasoning behind it never happened.
A six-month pilot of the Abridge scribe with 48 internal medicine residents, published in JGIM, named the risk directly: trainees becoming passive editors rather than active authors of care. The authors proposed guardrails tied to ACGME Core Competencies, including baseline documentation skills before scribe access and structured critique of AI-generated notes.
The practical fix is boring and it works. Make the resident draft the assessment and plan from memory first, then compare it against the scribe's version. The delta between the two is the teaching moment. Skip that step and the chart looks the same whether the resident thought carefully or didn't.
The Attending Who Can No Longer See the Thinking
Supervisors have usually inferred a trainee's reasoning from the written note, the oral presentation, and the differential. When the note has been polished by software, that signal degrades. A weak thinker and a strong one can produce charts you can't tell apart.
This is where the broader industry conversation lands squarely on medicine. Firms lose their junior training loop unless agents are built into the firm's own systems and review process, and hospitals face the same design problem.
If the scribe lives outside the teaching workflow, assessment goes blind. If it lives inside it, the review itself becomes the lesson.
The Early-Career Clinician Who Never Builds the Reps
Clinical reasoning is a muscle. It gets built through hundreds of iterations of pulling a messy encounter into a prioritized problem list. A widely cited Academic Medicine paper has argued that clinical reasoning deserves recognition as a core competency in its own right, precisely because weakness here drives diagnostic error downstream.
The uncomfortable part is that the reps have to happen somewhere. A handful of programs are trying structured substitutes:
- Pre-scribe drafts. The resident writes a one-paragraph assessment before the AI note is generated, and both go to the attending.
- Reasoning disclosures. A short addendum noting what the trainee accepted from the scribe, what they changed, and why.
- Chart-stimulated recall. The attending picks a note at random and asks the resident to defend each clinical decision without looking at it.
Programs Bolt the Scribe On Instead of Building It In
The failure mode worth avoiding is treating the scribe as an IT procurement decision that happens to the training program. Pick the vendor for throughput, plug it into the EHR, change nothing about how notes are reviewed, and the educational cost shows up two or three years later in weaker clinical reasoning at graduation.
The programs that keep their bench are doing the opposite. They are specifying what the scribe is allowed to generate for trainees, instrumenting the review step so faculty can see where the resident intervened, and handling the AI output as a teaching artifact rather than a billing document. The chart gets written either way. Whether the next generation of clinicians can still think through a hard case is the open question.