Descriptions:
As AI coding agents proliferate, developers face a practical problem: Claude Code and OpenAI’s Codex use different configuration conventions, meaning a project optimized for one tool won’t automatically work with the other. In this short tutorial, Nate Herk walks through exactly how to bridge the gap, showing a side-by-side comparison of both agents running against the same project and mapping every structural difference between them.
The key distinctions come down to naming and folder layout. Claude Code reads from a CLAUDE.md file and a .claude directory—containing agents (Markdown), skills, and settings—while Codex looks for an agents.md file, a .codex directory for config and agent definitions (TOML), and a separate .agents folder for skills. Shared knowledge files—reference docs, brand assets, project wikis—require no changes and work identically across both tools. The migration work is limited to duplicating or renaming config files and converting agent definitions from Markdown to TOML format.
Herk’s recommended shortcut is to prompt Codex directly: instruct it to read the existing CLAUDE.md, research its own documentation, and auto-generate the required agents.md and .codex config. The video includes a downloadable cheat sheet covering the full mapping between both systems, along with a list of five common beginner mistakes when switching between coding agents.
📺 Source: Nate Herk | AI Automation · Published May 18, 2026
🏷️ Format: Tutorial Demo
