# Zeteo is taking you through this one.
# Free, and it runs right here. No signup.
$ learn OpenAI Codex
Learn OpenAI Codex in 5 minutes.
Codex is OpenAI's software engineering agent: one model you can run in your terminal, in your IDE, or as cloud tasks that come back as pull requests. This lesson drills picking the right surface, setting approval modes, and writing an AGENTS.md that sticks.
# 5 minutes, no install
# interactive, not a video
# Lesson not loading? Open it in a new tab. It is the same free lesson either way.
Run it local, or send it to the cloud?
Codex is one agent with two homes, and the choice decides how much you watch it work. Local runs keep you in the loop; cloud tasks run unattended and return as diffs.
The CLI
Type codex inside a repo and describe the change. It reads your code, proposes edits and commands, and works right in the terminal where you can watch every step. Reach for it when you want to steer.
Delegated tasks
Hand Codex a task from ChatGPT and it works in an isolated cloud sandbox, then hands back a diff you can turn into a pull request. Right for well scoped jobs you don't need to supervise, and several can run in parallel.
Standing instructions
A file in your repo telling Codex how to build, test, and behave in this project. Write it once and every future run starts out already knowing your conventions.
What you will do in the lesson
You answer and try things. Nobody demonstrates at you for five minutes.
- Sort tasks into local and cloud. You get a handful of realistic engineering tasks and decide which belong in the CLI and which are safe to delegate. Then you see which choices backfire.
- Write a brief that survives without you. A cloud task can't ask follow-up questions mid-run, so you'll practice packing the goal, the constraints, and the definition of done into one message.
- Choose an approval mode on purpose. Codex can ask before every command or run freely inside its sandbox. You'll pick a mode for a risky task and justify it.
- Draft an AGENTS.md. You'll write the three or four lines that save the agent from rediscovering your test command and lint rules on every run.
- Read the diff like a reviewer. The lesson ends with a Codex diff that is mostly right. Your job is to find the part that isn't before you would merge it.
Before you start
Is Codex just ChatGPT writing code in a different window?
No. ChatGPT answers you in a conversation; Codex works in a real environment with your repository, running commands, editing files, and checking its own work. The output is a diff, not a snippet to paste.
Do I need to install the Codex CLI for this lesson?
No. The lesson runs in the frame on this page, with no signup and no download. Installing the CLI afterwards is optional if you want to try it on your own repo.
Can Codex run commands on my machine without asking?
Only if you tell it to. Approval modes control whether it asks before each command or works autonomously, and autonomous runs happen inside a sandbox with limits you set. The lesson covers when each mode is the sensible choice.
I already use Claude Code. Why look at Codex?
They overlap a lot in the terminal, and the habits transfer. Codex adds the cloud-task side: fire off several isolated runs and review the diffs later, which suits a different kind of work than an interactive session. Knowing both tells you which loop fits which task.
Other tools people learn next
Same format, same five minutes, still free.
Learners from these organizations are already learning on 
