OpenAI for Coding — Overview

The three OpenAI coding surfaces

There isn't one "OpenAI coding tool" — there are three surfaces that share models but suit different jobs. Picking the right one is most of the battle.

Codex CLI at a glance

The Codex CLI is a single binary you install once, sign into, and then run inside any project directory. Four things get you productive: install it, launch it, sign in, and learn the one command that makes it scriptable.

1 · Install

Pick whichever matches your setup — they install the same tool.

curl -fsSL https://chatgpt.com/codex/install.sh | sh

irm https://chatgpt.com/codex/install.ps1 | iex

npm install -g @openai/codex

brew install --cask codex

2 · Launch it in your project

cd into a repository and run codex. Just like Claude Code, it starts an interactive session scoped to that directory — it reads the files there and proposes edits and commands you approve.

cd my-project
codex

3 · Sign in

On first run Codex asks you to authenticate. The simplest path is "Sign in with ChatGPT" using a paid ChatGPT plan (Plus / Pro / Business / Edu / Enterprise) — no API key to manage. Alternatively you can authenticate with an API key if you'd rather pay per use or run in an environment without a browser login.

4 · Learn codex exec

The interactive codex session is where you'll spend most of your time, but the command that unlocks automation is codex exec: a non-interactive (headless) mode that runs a single instruction and exits, printing its result. It's the piece you drop into scripts and CI.

codex exec "run the test suite and fix any failing unit tests"

A tour of different usage

Beyond "launch and chat," a handful of subcommands and flags cover most real workflows. Each of these is copy-pasteable — try them in a scratch repo.

codex resume

codex cloud

codex -m <model-name> "refactor the auth module and explain the trade-offs"

codex --image ./bug.png "the layout breaks like this on mobile — find and fix the cause"

codex --search "upgrade us to the latest stable version of this framework"

codex --full-auto "add a README section documenting the config options"

For unattended runs, codex exec is the star. Here it is inside a CI job that runs on every push and lets the agent fix failing tests headlessly:

# .github/workflows/codex.yml (illustrative)
name: codex-autofix
on: [push]
jobs:
  autofix:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - run: npm install -g @openai/codex
      - run: codex exec "run the test suite; if tests fail, fix them and summarize what changed"
        env:
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}

Slash commands and project instructions

Inside an interactive session, slash commands manage the session itself — familiar territory if you know Claude Code's / menu:

Just like CLAUDE.md, AGENTS.md is a plain Markdown file at the root of your repo where you write standing instructions — conventions, how to run tests, what to avoid — so you don't repeat them every session. Codex reads it automatically.

How Codex CLI maps to Claude Code

Both are agentic terminal tools: you run them inside a repo, they read and edit files, run commands with your approval, and take standing instructions from a Markdown file. If you know one, this table gets you productive in the other quickly.

When to use which surface

Go deeper