AI agents that spawn other AI agents

A research agent gets a complex prompt. Instead of trying to do everything itself, it spawns three sub-agents in three new containers: one to scrape sources, one to summarize each source, one to cross-check claims against each other. Each sub-agent is a URL. The parent agent calls them like functions, gets back results, decides what to do next.

The sub-agents can spawn their own sub-agents. The fact-checker might spawn five parallel verifiers, one per claim. Each runs in its own isolated environment. None can corrupt the others. All speak HTTP to each other.

There's no orchestrator framework. There's no message bus to configure. There's no shared state to synchronize. The only thing that limits how many agents you run in parallel is how much you want to spend: idle containers cost nothing.

Each agent is a peer with a URL. They call each other the way any HTTP service calls any other HTTP service.