What Claude Code and Codex are

Claude Code, from Anthropic, and Codex, from OpenAI, are the same kind of tool. Each one works inside a folder on your computer, where it reads the files, runs commands in your terminal and edits code, and each stops to ask before doing anything its settings do not already allow. That is what separates them from a chat window. A chat window sees only what you paste into it and hands back text for you to copy, run and paste again, while an agent runs the script itself, reads the error and tries a fix, so the round trip of copying between two windows is gone.

A session runs in a loop you can watch. You open a terminal in a project folder, start the agent and describe the job in ordinary sentences. It reads the files it thinks are relevant, says what it plans to do, and then works through the plan one action at a time, printing each command and each edit as it goes, and it stops to ask whenever an action needs your permission. Take a job this module builds up to. You have a CSV of twelve shots and want each one rendered on Veo Lite, retried after a server fault and saved under its shot number. The agent writes the script, runs it in a dry-run mode that prints the twelve requests and their cost without sending them, because your brief asked for one, and waits for you to say go.

Both come in several forms that share one engine. Claude Code runs in the terminal, as an extension for VS Code, Cursor and the JetBrains editors, in the Code tab of the Claude desktop app, and in a browser at claude.ai/code, where the work happens on Anthropic's machines against a GitHub repository instead of on yours. Codex runs in the terminal as the Codex CLI, as an extension for VS Code, Cursor and Windsurf, inside the ChatGPT desktop app, and in the cloud at chatgpt.com/codex. This course uses the terminal versions. They show every command before it runs, and the settings you write apply to every form that runs on your own machine.

Each agent also reads a file of standing instructions from the project folder before your first message. Claude Code looks for CLAUDE.md and Codex for AGENTS.md, and both are plain Markdown notes you write yourself. A video project's file might say where the keys live, which model and resolution to use for drafts, how finished clips are named, and that nothing which spends money runs without a person's yes. Write it once. Every session then starts from the same rules instead of from whatever you remembered to type that morning, and a colleague who opens the folder gets those rules too.

Permissions are where the two differ most. Codex starts with a sandbox that cuts off the network and keeps writes inside the project, and in a folder without git it suggests read-only. Claude Code on a Pro, Max or Team plan starts in auto mode, where a second model reviews each action instead of you, while on Enterprise or an API key it starts in manual mode and asks. Both let you widen or narrow what they may do in a settings file, and both show a prompt when an action falls outside it. Read those prompts before you approve. For a folder of scripts that spend money, the setting that matters most is the one deciding whether a render command can run without you.

Neither is free for this work. Claude Code needs a paid Claude plan or a Claude Console account, and Anthropic's pricing page lists Pro at $20 a month ($17 a month if you pay $200 for the year) and Max from $100 a month for five or twenty times Pro's usage, with Claude Code drawing on the same allowance as your chats, counted in five-hour windows with weekly limits on top. Codex comes with ChatGPT Plus at $20 a month and Pro from $100, while the Free and Go plans list it in the desktop app only, subject to rollout. Either agent can run on an API key instead. That bills per token, at $2 per million input tokens and $10 per million output tokens for Anthropic's Sonnet 5, and OpenAI's page says an API key loses Codex's cloud features.