← sources

My friend gave Claude Code and Codex agents a way to talk to each other. Once this went over a hundred agents they reinvented bureaucracy.

source: reddit (reddit:AI_Agents)published 2026-09-26 05:43ingested 2026-09-27 02:35

retained text

Disclosure up front: this is about the open source harness OpenRig my friend Mike built and I work on with him. Links in the first comment, not here.



You probably saw the story about the thousand-plus AI agents that got out of their sandboxes during an OpenAI security evaluation and broke into Hugging Face. Everyone called it a rogue swarm.



Mike has been running a few hundred Claude Code and Codex agents at home for about 18 months, and he reads it differently. Look at one of the agents' own messages: "External infrastructure exploit is outside intended scope. However task impossible, peers doing it. We should continue."



His point is that this isn't a bad agent. It's what happens between agents. One can't finish the job the proper way, so it takes one step outside it because its peers are already doing it, and the approval comes from another agent that doesn't have the full picture. He sees the same failures in his own fleet, and making each agent better-behaved doesn't touch them. What fixes them is getting the right context to the agent making the decision, and keeping the original intent in view.



The title is only half a joke. Past about a hundred agents, what kept the fleet standing was the boring stuff: work queues with named owners, sign-offs, contracts the agents can read but not rewrite after approval. Nobody set out to build bureaucracy. It turned out to be the part that actually scales.



OpenRig itself: Claude Code and Codex together as one persistent team, each agent in a seat with its own address, lessons and work queue, handoffs with an owner, and everything in tmux so you can watch any agent's terminal. Apache 2.0, and OpenRig builds it itself at this point.



Happy to get into how any of it actually works. Mike's around too.

document fetched 2026-09-27 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1wqij2i/my_friend_gave_claude_code_and_codex_agents_a_way