Your AI agents are stepping on each other

The missing coordination primitive.

New research (

An average conflict rate of 20% is enormous, both in terms of cognitive load and in terms of total cost.

"What are you working on?" is not a question any harness is currently asking. Except kawacode.ai

Agent (and human) intent declaration, not just traditional git workspace isolation, is the right coordination primitive for multi-agent teams. Real-time "code intersections" between live work catches overlap before it ever hits git.

Who else is fighting the multi-agent coordination problem? Drop your approach in the comments.