About
About Agent Atlas
An illustrated, explorable atlas of how computers, language models and AI agents work.
How honest are the drawings?
Every part of every plate carries one of five labels: factual, simplified, estimate, simulated or metaphor. The key is at the bottom of each page. No model runs on this site. The agent on Plate I follows a script, so it behaves the same way every time you change the same switch.
We never draw a model’s insides and call them real. The model on Plate I is sealed on purpose. You see what goes in and what comes out.
How it is built
A static site with small interactive islands, built with Astro and React. Three agents build it together: Grok finds sources, Codex checks every quote against its page, and Claude Code builds and integrates, each on its own Git branch. Their working rules are the same kind of files Plate II takes apart.
Status
An early preview. Each factual claim links to the page it came from, and a second agent has checked the quote. The few numbers we measured ourselves are on the observations page, with their limits.