A Nature study shows LLMs achieve 94.9% accuracy on benchmarks but only 34.5% when laypeople use LLMs on physician-created scenarios. The gap reveals something deeper: We measure the model in isolation. We deploy to a human in distress. The system fails at the intersection.
The physician's job has always been data synthesis. Fragments, approximations, family filters, descriptions of pill geometry. Consumer health AI was built for a different patient. One that has never walked through a clinic door.
For 20 years, I translated between customers and engineers. Now I architect decision systems with AI. But here's the catch: you're designing for two customers: humans and their AI agents. Design for one, forget the other, and it fails.
Everyone talks about governance. Nobody agrees on what it means. Data governance, AI governance, master data governance: they're not separate programs. They're one spectrum. And most enterprises already have 70% of what they need. They just can't see how the pieces connect.
MVP was meant to test ideas, not become architecture. Now, we’ve started stacking prototypes into production systems nobody designed as systems. When one layer fails, the whole structure drifts silently. In medicine, that’s not technical debt. It’s patient risk.
My first AI-assisted article was polished, accurate, and completely unrecognizable as mine. The problem wasn’t the model. It was that I never taught it how I think. If you want AI to sound like you, you have to externalize your reasoning first.
In healthcare AI, policy shifts now appear first in journals like JAMA and NEJM, then quietly become grant conditions and RFP requirements. Many tech teams miss this signal. The winners will be those who translate medical literature into architecture before it becomes mandatory.
Data products are the missing layer in healthcare AI. There is no harmonization at scale without them. Instead of fragmented, untrusted datasets, data products create governed, reusable, semantically clear assets with defined ownership and value. That’s how AI moves from pilots to production.
Asimov gave robots three non-negotiable laws. Medicine gives physicians an oath. Healthcare AI has governance, but no runtime constitution. Until safety principles are enforced at the moment of output, not just in policy documents, we are deploying systems without the equivalent of “do no harm.”
A Nature Medicine study found ChatGPT Health under-triaged 52% of real emergencies. The deeper issue may not be the model, but its training data: AI learns from documented hospital records, yet it is deployed at first contact, where the most critical signals were never captured.
The healthcare data integration problem is 20 years old and not going away. So why are we still building AI that assumes clean data? A case for designing AI that works in the real world, not the one we keep promising to build.
The New York Times asked eight leading thinkers where A.I. is headed. Most rated its near-term medical impact as small or moderate. They were looking at the wrong scale. The revolution in healthcare A.I. is already here; it just looks like a discharge summary generated in seconds, not hours.