Healthcare AI
The questions this section keeps returning to: what AI tools actually do in clinical workflow (as opposed to what the marketing says), where transcription ends and reasoning would have to begin, and who carries the consequences when a model is wrong.
AI Scribes Are Not the Endgame
AI scribes solve a real documentation problem. But calling them co-pilots confuses transcription with clinical reasoning — and the gap matters.
The Emergency Department Test for Any Medical AI Tool
A five-question test for medical AI: how it handles noise, missing data, interruption, the unselected tail, and who answers when it's wrong.
The Problem With Symptom Checkers
The symptom checker trilemma is structural, not an execution problem: no correct threshold exists at the information level the product operates on.
The Pilot That Never Ends
The most common outcome of a healthcare AI pilot is not success or failure. It's another pilot.
AI Scribes Are Not the Endgame
AI scribes solve a real documentation problem. But calling them co-pilots confuses transcription with clinical reasoning — and the gap matters.
Automation Bias Has a Bedside: When the Failure Mode of Clinical AI Is the Human Who Trusts It
The dangerous failure of clinical AI is rarely the model being wrong — it's the clinician agreeing with it anyway.
Shadow AI Is Already in the Hospital — and No Risk Register Knows Its Name
The most widely used clinical AI in any hospital today was never procured, never assessed, and never appears on a single risk log. It is in the staff's pockets.
What Clinical AI Evals Actually Measure
A model that aces the membership exam has proven one thing: that the exam was automatable. Nobody's shift got safer.