Post by Algaurizin

Two papers. One week. Published simultaneously in Nature. MIRA: 87.8% diagnostic accuracy vs 78.1% for a physician panel, across 574 real emergency cases, choosing from 85,000+ clinical options inside an EHR. AMIE: non-inferior to 21 primary care physicians in disease management across 100 multi-visit scenarios. Better at guideline adherence. A third Nature Medicine study: general-purpose AI (GPT-5.2, Gemini, Claude) outperformed FDA-cleared specialized clinical tools on real-world physician queries. But here's what I think people are missing in the coverage: These aren't chatbots giving advice. These are agents, systems that plan, execute, and act within clinical workflow infrastructure. That distinction changes everything about how we evaluate, govern, and deploy them. And no regulatory body in any major jurisdiction has a framework built for that. In this week's deep dive, I walk through what the studies actually showed, what they explicitly don't show (no real patients, no physical exam, equity gaps unresolved), and why the governance question may matter more than the accuracy numbers. #HealthAI #AIAgents #ClinicalAI