“Agents only start working on tasks that aren’t blocked, so execution unfolds naturally and optimally in parallel for this DAG (a sequence of execution steps).”

— Alex Kotliarskyi, Victor Zhu and Zach Brock, OpenAI · Turning our issue tracker into an agent orchestrator

Quote · openai.com

From “An open-source spec for Codex orchestration: Symphony”, found 2026-10-14.

Take

The unit being scheduled is the ticket and its dependencies, not the agent session. Agents also file follow-up issues that are evaluated and scheduled later, and crashed or stalled agents are restarted.