Subagents are now available in Codex. You can accelerate your workflow by spinning up specialized agents to: ⢠Keep your main context window clean ⢠Tackle different parts of a task in parallel ⢠Steer individual agents as work unfolds https://t.co/QJC2ZYtYcA
OpenAI Codex Now Runs Parallel Subagents for Multi-Agent Workflows
Ā· Updated
Codex now supports subagent workflows ā spawn specialized agents that run in parallel and consolidate results in one response. Three built-in agents ship by default:
default, worker, and explorer. Subagents surface in the Codex app and CLI. Codex only spawns subagents when explicitly asked, and each inherits the parent session's sandbox policy.Parallel execution addresses a core challenge in agentic coding ā context pollution. When each agent focuses on one concern, the main thread stays clean. Custom agents are defined via TOML files with their own model, sandbox mode, and MCP server configuration, letting teams build purpose-built agent crews for their workflows.
Spin up a parallel PR review ā ask Codex to dedicate one agent to security, one to code quality, one to test flakiness ā and get a consolidated result. Codex handles the orchestration: spawning, routing follow-up instructions, waiting for results, and closing agent threads.
OpenAI Developers
@OpenAIDevs
257retweets
View on X


