Use subagents and custom agents in Codex

2026-03-16 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

OpenAI Codex has released subagents in general availability, a feature previously in preview. These subagents, similar to those in Claude Code, include default roles like "explorer," "worker," and "default." While the distinction between "worker" and "default" is not explicitly clear, "worker" appears designed for parallel execution of numerous small tasks. Codex also enables users to define custom agents via TOML files in `~/.codex/agents/`. These custom agents can be configured with specific instructions and assigned to particular models, such as `gpt-5.3-codex-spark` for enhanced speed. The documentation provides an example prompt demonstrating how to orchestrate custom agents like `browser_debugger`, `code_mapper`, and `ui_fixer` to diagnose and resolve software issues.

Key takeaway

For AI Architects designing automated development workflows, you should explore OpenAI Codex's subagent and custom agent capabilities. Defining custom agents with specific models like `gpt-5.3-codex-spark` can optimize performance for specialized tasks, while orchestrating multiple agents in a prompt allows for complex problem-solving, such as debugging and code remediation. This approach offers enhanced modularity and control over AI-driven development processes.

Key insights

OpenAI Codex now offers subagents and custom agents for modular, specialized AI-driven development tasks.

Principles

• Modular agents enhance task specialization
• Custom agents allow model-specific tuning

Method

Define custom agents in TOML files with specific instructions and model assignments, then orchestrate them by name within prompts to tackle complex development problems.

In practice

• Use "worker" subagent for parallel tasks
• Assign `gpt-5.3-codex-spark` for speed
• Orchestrate agents for debugging workflows

Topics

• OpenAI Codex
• AI Subagents
• Custom AI Agents
• AI Coding Assistants
• Multi-Agent Systems

Best for: AI Architect, AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• 🎧 How OpenAI’s Codex Team Uses Their Coding Agent — OpenAI's Codex is expanding its user base and capabilities, emphasizing speed and a dedicated GUI for enhanced developer experience.
• Subagents have arrived in Gemini CLI — Gemini CLI subagents delegate complex tasks to specialized, isolated agents, preserving primary context and enabling parallel execution.
• How Endava builds an agentic organization with Codex — Codifying senior expertise into AI agents like Codex enables organizations to scale knowledge, accelerate delivery, and elevate junior talent.
• Building a Team of AI Agents: Roles, Feedback, & Teamwork Explained — Complex AI agent tasks benefit from a collaborative team structure with specialized subagent roles.
• openai / openai-agents-python — The OpenAI Agents SDK enables building flexible, multi-agent LLM workflows with robust control and extensibility.
• TAI #205: Codex Special Edition; Interviewing Romain Huet, OpenAI's Head of Developer Experience — The definition of a developer is changing, with AI agents enabling anyone to build complex systems through "agentic delegation."

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.