CAMMELOT
What happens when AI runs a Dutch town's healthcare?
45 pixel citizens · 3 GPs · 1 hospital · 100 simulation runs · Real Dutch data
The Story
Chapter 0
“I work at Microsoft. You should assume I’m biased toward AI. That’s exactly why I built something that could prove me wrong.”
A simulated Dutch town. 45 citizens. Real data. Two hundred simulation runs. Here’s why I built a machine that could prove me wrong — and in several important ways, did.
Read chapter →
Technical Deep Dives
Chapter 1
“Every agent publishes a card. Every action writes a FHIR resource. The mesh self-organizes.”
Agent Cards, FHIR memory stores, and 200 lines of JavaScript. How a protocol-first architecture makes the IZA’s €2.8 billion integration problem look very different.
Read chapter →
Chapter 2
“Data mesh promised decentralized ownership. It delivered decentralized abandonment.”
Why data mesh fails in healthcare, and what an agent-native architecture does differently. Discovery over integration. Context travels with action.
Read chapter →
Research · Coming Soon
Series 3
Mortality, burnout, costs, fairness. Every finding — including the ones that embarrass the AI-optimist position.
Series 4
What this means for the IZA, the zorginfarct, and 301,000 missing healthcare workers.
Preview
Chapter 3
“I ran 100 simulations. I have numbers. Some confirmed what I expected. Others didn’t.”
Results on mortality, burnout, costs, and fairness. Charts, effect sizes, and p-values. Coming week by week.
See what’s next →