OpenAI has officially moved subagents into general availability for Codex, signaling a major shift toward "agentic engineering" where AI handles complex coding tasks by breaking them down into smaller, specialized roles. This update introduces three default subagent types—"explorer," "worker," and "default"—which allow the system to navigate file structures and execute parallel tasks more efficiently than a single linear model.

Beyond the built-in defaults, developers can now craft custom agents by creating configuration files in the TOML format. These files allow users to assign specific instructions and choose high-performance models, such as gpt-5.3-codex-spark, to optimize for speed. By defining distinct personas like a "browser_debugger" or a "ui_fixer," the system can autonomously investigate bugs, trace code paths, and implement fixes without constant human intervention.

This move aligns OpenAI with a growing industry trend, as subagent architectures are becoming standard across major coding tools. Competitors like Anthropic’s Claude Code, Google’s Gemini CLI, and Mistral Vibe are all adopting similar patterns. This widespread adoption suggests that the future of software development will rely less on chatting with a single AI and more on orchestrating a fleet of specialized digital assistants.

OpenAI has officially moved subagents into general availability for Codex, signaling a major shift toward "agentic engineering" where AI handles complex coding tasks by breaking them down into smaller, specialized roles. This update introduces three default subagent types—"explorer," "worker," and "default"—which allow the system to navigate file structures and execute parallel tasks more efficiently than a single linear model.

Beyond the built-in defaults, developers can now craft custom agents by creating configuration files in the TOML format. These files allow users to assign specific instructions and choose high-performance models, such as gpt-5.3-codex-spark, to optimize for speed. By defining distinct personas like a "browser_debugger" or a "ui_fixer," the system can autonomously investigate bugs, trace code paths, and implement fixes without constant human intervention.

This move aligns OpenAI with a growing industry trend, as subagent architectures are becoming standard across major coding tools. Competitors like Anthropic’s Claude Code, Google’s Gemini CLI, and Mistral Vibe are all adopting similar patterns. This widespread adoption suggests that the future of software development will rely less on chatting with a single AI and more on orchestrating a fleet of specialized digital assistants.