Overview
OpenAI Codex has launched subagents in general availability, allowing developers to coordinate multiple specialized AI agents for complex coding tasks. This represents the mainstream adoption of multi-agent workflows in AI-powered development environments.
Key Facts
- Subagents now generally available in OpenAI Codex - developers can coordinate specialized AI agents instead of relying on single-agent solutions
- Default subagents include explorer, worker, and default agents - complex debugging tasks can be broken down and parallelized across specialized agents
- Custom agents can be defined as TOML files with specific instructions and models - teams can create domain-specific AI workflows tailored to their codebase
- Example workflow: browser_debugger reproduces bugs, code_mapper traces code paths, ui_fixer implements fixes - end-to-end problem solving without manual coordination between tools
- Subagent pattern now supported across major platforms (Claude Code, Gemini CLI, Mistral Vibe, VS Code, Cursor) - multi-agent development is becoming the industry standard
Why It Matters
This signals a fundamental shift from single AI assistants to orchestrated AI teams for software development, potentially transforming how complex coding problems are approached and solved.