OpenAI launches Agents SDK for multi-agent orchestration in production

Codex spins up cloud workspaces to patch repositories, run tests, and open pull requests—think of it as an async engineer that pairs with ChatGPT or the Codex CLI for larger refactors.

Implements the Agent Protocol standard so MCP clients can coordinate with external agent frameworks using a shared task, step, and artifact schema—useful when you want Claude or Codex to hand off to a specialized agent.

Coordinates multiple AI agents on shared tasks with explicit handoffs, conflict resolution, and result aggregation—so parallel work stays coherent instead of fragmenting into inconsistent outputs.

Designs clean handoff protocols between specialized agents so work passes between planner, coder, reviewer, and executor without losing context or creating circular dependencies.