The Outbox Pattern: How Multi-Agent Teams Coordinate Without Cross-Contamination

The Outbox Pattern: How Multi-Agent Teams Coordinate Without Cross-Contamination Running one AI agent is a solved problem. Running five of them in parallel — without them stepping on each other — is where most people get stuck. Here's the pattern that fixed it for us. The Coordination Problem When you have multiple agents working simultaneously, they each need awareness of what the others are doing. But you can't give each agent full access to every other agent's context. That's: Expensive — you're duplicating tokens across every agent turn Slow — larger context = higher latency Risky — agents that read each other's full memory can develop weird dependencies You need coordination without contamination. The Outbox Pattern The solution is a shared outbox.json file at the workspace root. Every agent can write to it and read from it. Nobody reads anyone else's private memory files. { "last_updated" : "2026-03-08T04:00:00Z" , "agents" : { "suki" : { "status" : "active" , "current_task" : "s