Lesson 0 / 7 · ~8 minutes
Codex: an agent,
not a chat
You know ChatGPT: you type a question, you get an answer. Codex is a different animal. It doesn't advise you on how to do things. it does them. It reads your files, writes code, runs it, tests it and hands you a finished result to review.
While classic vibecoding has you sitting in a conversation with AI, building the thing together step by step, Codex is a delegate: you brief a task, it goes off to work, and you come back to a finished result. It will happily work on three things at once, and happily while you're at lunch.
The delegation loop
Brief the task so the agent can handle it without you. The agent works for minutes, sometimes half an hour. Review the result: the diff, the logs, the tests. Then decide: accept, adjust, or re-brief differently.
Where Codex runs
- The desktop app: since July 2026, Codex is part of the ChatGPT app for Windows and Mac. This is the most comfortable way to start.
- The cloud: tasks run in an isolated environment on OpenAI's servers with a copy of your project. Your computer can sleep meanwhile.
- Mobile: you brief and review tasks straight from your phone, and via Codex Remote you can even drive a paired computer from afar.
- The terminal (CLI) for command-line lovers. You'll see it a few times in this course's replays.
What it's good for
The same rule as all vibecoding: brilliant for websites, calculators, internal tools, scripts and changes to existing code, things whose results you can verify. Not for critical systems without experienced oversight. And with Codex it applies doubly: the better you can describe the task up front, the more you get out of delegation.
Test yourself: myth or reality?
Five claims about Codex. Click what you think.
“Codex works even while I sleep.”
Reality. Cloud tasks run on OpenAI's servers in an isolated environment. Brief in the evening, review in the morning.
“I need an API key and to pay per token.”
Myth. Codex is part of the regular ChatGPT subscription, no API keys. How many tasks you get depends on your plan.
“Codex is just better code completion for programmers.”
Myth. It's a full agent: it reads the project, writes, runs, tests and returns a finished solution. You write the brief in plain sentences.
“I can brief a task from my phone on the way back from lunch.”
Reality. From the mobile ChatGPT app you can create, watch and fork tasks. You review the result when you're back at your desk.
“If the tests passed, I can accept the result without looking.”
Myth. Reviewing the diff is your half of the job. Tests verify nothing broke, but whether the app does what you want, only you can tell.
Summary
- Codex = a delegate: BRIEF → AGENT WORKS → REVIEW.
- It runs in the ChatGPT app, the cloud, on mobile and in the terminal.
- It runs on a regular ChatGPT subscription, no API keys.
- Reviewing the result is your job. The agent does, you decide.