Dr. Friedman is a physician turned product manager with 20+ years building enterprise software and leading digital transformation. He writes about the intersection of technology, human behavior, and healthcare, where solutions directly impact lives.
AI doesn't hallucinate randomly. It hallucinates in patterns. A new paper from MIT, Harvard, and Google maps five of them, each with a distinct failure mode and a distinct way of passing review. The taxonomy was built for medicine. The failure modes are universal.
OpenAI had one of the best models in the world. Walmart still fired it. Not because the model was weak. Because it had no idea what it was working with.
The people most affected by your AI product are almost never in the room when you build it. They don't appear in your analytics. They weren't in the meeting when the roadmap was approved. They will never give you feedback. You owe them something anyway
AI compresses knowledge. It does not compress experience. The junior PM in your next interview has access to the same tools you do. What separates you now is judgment. There is no course for that. There never was.
Healthcare AI job postings keep asking for unicorns: 14 years of experience in a field that is three years old, deep AI expertise, deep clinical expertise, deep product expertise. That person does not exist. And even if they did, they would not fix the real problem.
In healthcare, AI can already predict. The hard problem is reasoning, and reasoning depends entirely on context. The physician who takes a complete history, examines carefully, and aggregates everything from the EHR to the nursing home fax is not being thorough. They are building the product.
People don't buy a drill. They buy a hole. AI is the drill. Every major technology wave produces the same confusion. SQL became infrastructure. Java became infrastructure. Cloud became a checkbox. AI is doing the same thing. We are in the expensive middle of that arc right now.
Healthcare is slow, regulated, fragmented, and over-documented. That's exactly why it's the best AI proving ground in the world. The field that everyone called a laggard has been running the most rigorous AI experiment for decades. The rest of us are just starting to catch up.
We evaluated AI health tools by removing the very mechanism that makes diagnosis work, measured what happened, and called it a safety problem. It is not just a safety problem. It is a design problem. Medicine knew this in 1975
Two studies this week, one from Google and one from Microsoft, are being celebrated as evidence that healthcare AI has arrived. Read them together and they reveal something more uncomfortable: AI is filling a gap that healthcare created long before any of us started building AI for health.
Twenty years of customer interviews, workshops, and journey maps. Then agentic AI arrived, and every framework I trusted turned out to share one assumption I had stopped noticing: that the human is always smarter than the tool. Here's what breaks when that stops being true.
The clinical reasoning pattern hasn't changed since Hippocrates. The vessels carrying it have, from paper charts to cloud EHRs. For fifty years, every vessel moved the data without understanding it. AI breaks that pattern for the first time. Not a better vessel. One that thinks.