The newest evidence on ambient AI scribes is genuinely encouraging. In a 2026 multisite JAMA study across five academic medical centres, adoption was associated with lower EHR and documentation time and a small increase in weekly visit volume.

I would absolutely pay attention to that if I were running a clinic. Documentation is real work, it spills outside scheduled hours, and reducing it matters.

But the unit of analysis matters.

A scribe can make the note easier without making the pathway easier. The patient may still wait for a referral to be triaged. A staff member may still chase missing information. A physician may still receive an inbox that is impossible to clear. The same clinic can be faster at documenting an encounter while still being slow at moving the patient through care.

Saving minutes inside one task is not the same thing as removing friction from the system.

That distinction is why I like this research. It makes the conversation more specific. We can stop arguing about whether AI “works” and start asking where it changes time, quality, capacity and cognitive load.

The outcome I would watch next is redistribution.

If documentation time falls, where does that time go? More visits? More patient conversation? More inbox work? Earlier departure? A different task that used to be deferred? Productivity is not one number, especially in care environments where every saved minute can be immediately absorbed by unmet demand.

A 2025 JAMA Network Open study similarly found lower EHR and note time among AI-scribe users, but no significant change in after-hours documentation, appointment length or monthly office-visit volume. That does not weaken the case for scribes. It makes the case more precise.

My implementation question is boring on purpose.

What does a normal Tuesday look like after the scribe goes live? Who reviews the generated note? What happens when it is wrong? Which phrases or sections get checked every time? What new exception workflow appears? Who owns adoption, training and monitoring?

Those questions are less exciting than a demo. They are also where a technology becomes an operating model.

REFERENCES / FURTHER READING

  1. Shah et al. Changes in Clinician Time Expenditure and Visit Quantity With Adoption of Artificial Intelligence–Powered Scribes. JAMA. 2026. JAMA — AI Scribes and Clinician Time Expenditure ↗
  2. Pearlman et al. Use of an AI Scribe and Electronic Health Record Efficiency. JAMA Network Open. 2025. JAMA Network Open — AI Scribe and EHR Efficiency ↗
  3. Everson et al. Uptake of Generative AI Integrated With Electronic Health Records in US Hospitals. JAMA Network Open. 2025. JAMA Network Open — Generative AI Uptake in U.S. Hospitals ↗

These are personal research notes and commentary. I link the underlying sources so the evidence can be checked directly.