Codex 101: Getting Started (OpenAI's Answer to Claude Code)

Codex is OpenAI's version of the tool I use most: an AI agent that works directly inside a folder of files. If you've read the Claude 101, the one-line translation is that Codex is OpenAI's answer to Claude Code. And here's the part that makes this page worth five minutes even if you never open Codex: it's the living proof behind the most important architectural decision in my whole setup.

What Codex is

Codex is an AI agent built to work inside files and projects. Developers use it to write and fix software, and that's most of what you'll read about it elsewhere. But look underneath the developer marketing and the core ability is broader: point it at a folder, and it can read what's there, work across many files at once, and create and edit them toward a goal you set.

That's the same shape as Claude Code. Two rival companies looked at the future of AI tools and landed on the identical conclusion: the next step past the chat window is a worker with access to your actual files. When both giants build the same thing, that's not a feature. That's where the whole field is going.

Why a non-developer should care

My entire business lives in one folder of plain files: meeting notes, strategies, content plans, client histories, and the rules my AI runs on. Today I mostly point Claude Code at it. But nothing about my system depends on Claude, because it's just files. If I want to open the exact same folder in Codex tomorrow, or in some tool that doesn't exist yet, I can.

Play that forward, because this is the strategy underneath everything I teach:

• The "best tool" changes every few weeks in AI. Model leapfrogging is the norm now, not the exception.
• If your system lives inside one company's app, every leapfrog forces a choice: painful migration, or staying somewhere worse because leaving costs too much. That second option is how people end up resenting their tools.
• If your system is a folder of plain files, switching is nearly free. You keep the brain and swap the engine. Codex existing (and competing hard with Claude Code) is exactly why the engines keep getting better and cheaper for people whose files aren't hostages.

So even if you never touch it: structure your second brain so Codex COULD read it tomorrow, and you've future-proofed against every AI company's roadmap, including the one you currently love.

Getting started (if you want to actually try it)

1. You'll need a ChatGPT account. Codex ships as part of OpenAI's paid offerings, and exactly where it lives in the lineup keeps shifting; check openai.com/codex for the current entry point rather than trusting any screenshot-heavy tutorial older than a season.
2. Point it at a copy, not the real thing. Duplicate a small folder of notes and let it work there first. This is just good hygiene with any file-touching agent.
3. Give it a reading task before a writing task.

Read everything in this folder and tell me: what is this project,
how is it organized, and what's inconsistent or missing?

How well it answers tells you immediately whether it understands your structure, and the "inconsistent or missing" clause usually surfaces something genuinely useful.

4. Then one real writing task you can verify:

Create a summary file at the top of this folder: one paragraph per
subfolder describing what's inside, written plainly. Don't change
any existing files.

Constrained, checkable, and representative of real second-brain housekeeping.

5. Run the same two tasks in Claude Code on the same folder copy. Same files, same asks, side by side. That's the honest way to evaluate any AI tool: not demos and launch videos, your work.

How to think about Codex vs. Claude Code

They're the same species, so the comparison is mostly about everything around the tool:

• Ecosystem: if your world is ChatGPT (custom instructions, image work, voice capture), Codex keeps you in one subscription. If your stack runs on Claude connectors, scheduled tasks, and a Claude-trained rules system (mine does), Claude Code compounds with all of it.
• The work itself: for reading, organizing, and maintaining a folder of plain-text business files, both are capable. Differences show up at the developer end of the spectrum, which isn't where this audience lives.
• The wrong question: "which one wins?" The right question: "is my system structured so it doesn't matter?" Plain files, clean folders, a README on top. Then you get to choose fresh every year, from strength.

Mistakes to avoid

• Evaluating it on someone else's demo. Agents look magical in demos and honest on your own messy files. Test on yours.
• Letting either tool restructure your brain to its taste. The folder structure serves you and every future tool, not this quarter's favorite. You're the architect; agents are the crew.
• Paying for both out of FOMO. Pick the ecosystem you already live in. Revisit twice a year, not every launch day.
• Reading "Codex vs Claude Code" flame wars as strategy. The portable-files decision matters ten times more than the pick itself.

FAQ

What's the difference between Codex and ChatGPT? ChatGPT is a conversation. Codex is an agent that works inside folders and projects: reading, creating, and editing files toward a goal. Same company, different shape of tool.

Codex or Claude Code? Same species of tool. I use Claude Code daily because the rest of my stack lives in Claude, and my files would open in Codex just as well. Pick the ecosystem you already pay for, and structure your files so you could always switch.

Do I need Codex if I don't code? Not today. You need what it represents: proof that keeping your second brain in plain, portable files means every AI company's file agent is a possible engine for YOUR brain, forever.

Will these file agents replace the chat tools? They're additions, not replacements. Chat for thinking and drafting, agents for working across files. The Claude 101 maps the same split on the Claude side.