What agent orchestration is.

Agent orchestration is what happens when you turn one AI into a small team.

Instead of asking one chatbot to do everything, you split the work across a few smaller agents, each with one job. One drafts, one checks, one decides. They never talk to each other directly; a tiny program in the middle picks who runs next and what they get to see. In Three-Body Agent that middle program is about fifty lines of code and the three agents never share notes, only the queue between them.

Each agent has one job, like specialists in a hospital.

You would not ask one doctor to do surgery, fill prescriptions, and run X-rays. Same idea here: each agent gets a narrow role, like "write," "review," or "decide." Swap the AI behind a role and the rest of the team keeps working the same way, because the job description did not change. Claude Code's subagents work like this already.

When one agent messes up, the next one catches it.

AI makes confident mistakes all the time. If the same one is doing everything, the mistake gets baked in. If a separate agent reviews the work with a different goal in mind, the mistake gets logged and rejected. In Three-Body Agent one agent writes code, another tries to break it, and a third decides who is right; that argument is what stops bad code from shipping.