After the Prototype - agents, applied AI, and models put to work

Format: 15-minute flash talk with a live demo, or 45 minutes if you’d like the full version.

Most of an agent’s life is spent waiting on the model, about 85% of each turn in my tests. So once agents start spawning other agents, what really matters is what each one costs while it waits. I built Rui in Zig so that one process runs every agent’s loop, with their state kept as rows in SQLite. A hundred waiting agents take 7.4 MB. Codex app-server takes 210 MB and OpenCode 517 MB. I’ll show a live swarm, walk through where the memory goes in other harnesses, and talk about what broke when I ran 101 real agents on my laptop. It’s for anyone running agents or planning for swarms.

Links: https://github.com/DivyanshGolyan/rui and https://divyanshgolyan.com/posts/hundred-agents/