AI agents for medical secretarial work: what a randomized trial found about time saved and burnout

Medical secretarial work — scheduling, reminders, drafting the consultation note, updating the record — is one of the sectors where AI agents are furthest along, because the tasks are structured and the volume is high. It is also one of the few areas with an actual randomized controlled trial behind the claims, rather than only vendor demos, which makes it worth looking at the study rather than the marketing.

What the trial measured

A parallel three-arm randomized trial run from November 2024 to January 2025, A Randomized-Clinical Trial of Two Ambient Artificial Intelligence Scribes: Measuring Documentation Efficiency and Physician Burnout (posted as a 2025 preprint, not yet peer-reviewed), assigned physicians to one of two ambient AI note-drafting tools or a usual-care control group. One of the two tools produced a statistically significant 9.5% reduction in time spent on documentation compared to the control; the other showed no significant change. Physicians using either tool reported modest but measurable improvements in burnout and task-load scores compared to control, with higher usage correlating with a bigger benefit.

The honest read of that result

A 9.5% reduction in documentation time is real and worth having — it is also not the "half your paperwork disappears" claim often implied by product marketing, and the second tool in the same trial showed no measurable time benefit at all, which is the more useful finding: the technology does not perform uniformly, and a pilot on your own workflow tells you more than a vendor’s case study.

• →Pilot with a small group before a full rollout — the trial shows the two tools tested did not produce the same result, and yours may not either.
• →Keep the draft note under mandatory physician review before it enters the record — an ambient scribe drafts, it does not diagnose or certify.
• →Measure your own time-in-note and burnout indicators before and after rather than relying on published averages from a different setting.