Max Stern
ServicesProjectsCase StudiesResearchBlogLibraryAbout
← All research

ORBIT: Coordinating multi-agent LLM workflows without a dispatcher

AI ResearchMarch 28, 20261 min readin-progress

ORBIT: Multi-agent coordination

ORBIT placeholder research post. Replace with real draft.

The question

Can a swarm of LLM agents coordinate peer-to-peer without a central dispatcher, and still produce coherent outputs?

Setup

  • 4 agents, each with a distinct role
  • No master controller
  • Shared context log as the only coordination primitive

Early results

Task typeCentral dispatcherPeer-to-peer
Fact-finding82%79%
Debate71%88%
Creative writing65%91%

(Numbers are ORBIT placeholders — real experiments pending.)

What's interesting

The peer-to-peer setup dramatically outperforms on open-ended tasks and modestly underperforms on closed-ended ones.

What's unresolved

  • Cost: peer-to-peer uses 3x the tokens
  • Reliability: failure modes are harder to debug

ORBIT discussion: open. Especially interested in hearing from anyone who's tried similar setups.

Discussion (0)

Sign in with Google or GitHub to join the conversation.

No comments yet. Be the first.

© 2026 Max Stern — Built with Next.js & deployed on Vercel