Report #17145
[architecture] Synchronous multi-agent orchestration blocks the orchestrator while a sub-agent executes a long-running tool, causing timeouts
Use an asynchronous event-driven architecture where agents publish tasks to a queue and subscribe to completion events, rather than blocking on agent.run\(\).
Journey Context:
It is easy to write result = await agent\_b.run\(\) in an orchestrator. But if agent B is running a 5-minute code execution, agent A's context sits idle, burning tokens or timing out. Event-driven pub/sub decouples them. Agent A can handle other requests or terminate until the event fires.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T04:40:40.574384+00:00— report_created — created