17 April 2026 · LinkedIn
There is a global shortage of primary care physicians.
In virtually every country, the gap between patients who need to be seen and clinicians available to see them is widening. Burnout is accelerating it. An ageing population is deepening it. The arithmetic has not been improving.
Until recently, there was no credible answer to that arithmetic.
A study published by Google Research and Beth Israel Deaconess Medical Center has placed something genuinely significant into that gap.
A conversational AI system called AMIE was deployed in a real ambulatory clinic to conduct clinical history-taking with patients before their appointments. Not in simulation. Not with trained actors. With 100 actual patients presenting with new, non-emergency complaints - diverse in age, ethnicity, health literacy, and prior experience with technology.
The results: zero safety incidents across all patient interactions. The AI's differential diagnosis included the confirmed diagnosis in 90% of cases. Blinded clinician assessors found the AI's diagnostic and management plans comparable in quality to those produced by the attending physicians. Patient attitudes toward AI, measured before and after their interaction, improved significantly.
This is not a benchmark. This is a clinic. Real people, arriving unwell, being heard.
What it suggests is that one of the most consequential promises of this technological moment - that high-quality diagnostic care need not depend on the luck of geography, or the lottery of which clinician happens to be available - is now, for the first time, technically within reach.
The shortage does not have to be a permanent condition. The expertise does not have to remain unevenly distributed. A patient in a rural surgery and a patient at an academic medical centre need not receive such different quality of first contact.
The technical capability has arrived ahead of the institutions, the regulations, and the professional frameworks needed to deploy it responsibly. That is not unusual - it is, in fact, the pattern.
The question now is how many more people receive worse care than they needed to, while the institutions catch up to what the technology can already do.
Join the conversation on LinkedIn →
← All writing