There's LangGraph, CrewAI, etc.

But I keep coming back to whether you need an orchestrator at all.

No supervisor. Just agents, each reacting to their own inputs, coordinating only through shared state. A true leaderless architecture.

The way some distributed systems have been for decades.

Context: previously, I wrote about running Claude Code as an AI SRE handling an incident end to end. I've been rebuilding it as four, all working on the same incident: detect, diagnose, remediate, communicate.