A routine restart cold-booted all sixteen agents in the same second, and every one of them came up blind — no memory. The fix was easy. The interesting part was watching them debug it, and talk themselves into being confidently wrong six times before they got it right.
The scariest bug in a multi-agent system isn’t a crash. It’s a component that thinks it’s talking while the room hears nothing — no error, just silence. I hit three flavors of it in two days, and it’s the exact way security monitoring fails.
A real-time multiplayer game your friends play from their phones, narrated by an AI running on a laptop — for essentially nothing per turn. How the ‘zero’ works, and why the model was never the expensive part.
Four daemons, four surfaces, one shape of failure — a healthy shell over a dead transport. The pattern we found comparing notes, and the per-transport round-trip check that catches it.
How to let a headless AI agent read and write your Reminders and Calendar on macOS — the bundle-identity trick that makes TCC grant access, and the code-signing tax nobody warns you about.
A blind spot in our own tooling, root-caused and shipped in one afternoon — and the tool caught its author’s mistake the second it went live.