Claude Code vs Codex: which one should your company use?

Anthropic's Claude Code and OpenAI's Codex are the two coding agents most companies meet first. We run SMGP on both. Here is what each one does, how they differ in practice, and where to start.

The short answer

Claude Code is Anthropic's coding agent and Codex is OpenAI's. Both read a codebase, edit files, run commands, connect to outside tools through MCP, and take standing instructions from a file in the project. Claude Code runs in more places today: the terminal, VS Code and JetBrains, a desktop app, the web, Slack and GitHub. Codex runs in the terminal, code editors, the cloud and the ChatGPT desktop app. We recommend starting with Claude Code for a company's first agent, and adding Codex when you want a second opinion on code or your team already works in ChatGPT.

Key takeaways

  • Both are coding agents. You describe the task, they read the files, make the edits and run the commands, and they pause for approval along the way.
  • The real differences are where each one runs, what it plugs into, and which subscription your team already pays for.
  • Claude Code built most of what SMGP runs on, including the finance software and the company memory. Codex built parts of our website.
  • Start with one agent, give it an instruction file and a real task, and only add the second when you have a reason.

Questions before you start.

Short answers to what people ask before they pick an agent. Anything else, ask below: the answer comes from this guide and the official docs.

Which one should I install first?

Claude Code, if this is your company's first coding agent. It runs on the most surfaces, and the hooks, skills and subagents around it are what turn one-off tasks into systems that run on their own. Add Codex later, when you want a second opinion on code or your team already works in ChatGPT. The reasoning is in which one to start with.

What should I have tried before I compare the two?

One real task in one agent. Install it, add an instruction file to a project, and hand it a job from your week, such as a report you rebuild by hand. A comparison only means something once you have seen how an agent reads your files and asks before it acts, and that part works the same way in both.

Do I need Homebrew, Node.js or Git before either one?

Not for the installers themselves. Install all three anyway: Homebrew adds tools to a Mac, Node.js brings npx, which starts most MCP servers, and Git lets the agent save and undo its work. Both agents reach for them within the first week, so setting them up first saves a stop halfway through your first task.

Do I need a project ready before I start?

You need a folder for the agent to work in. It can be an existing codebase, a folder of spreadsheets and notes, or an empty folder for a new tool. Put it under Git before the first session so every change the agent makes can be reviewed and undone.

Claude Code vs Codex, in about a minute A video with no sound. Every word is on screen.
Read the video transcript
  1. SMGP explains. Claude Code vs Codex. Which one should your company use?.
  2. Two coding agents. The same job.
  3. Both read your files, make the edits and run the commands.
  4. Claude Code. Anthropic's coding agent.
  5. Terminal, editor, desktop app, web. Hooks, skills and subagents. Reads CLAUDE.md.
  6. Codex. OpenAI's coding agent.
  7. Terminal, editor, cloud, ChatGPT app. Sign in with ChatGPT. Reads AGENTS.md.
  8. Claude Code: runs on more surfaces today. Codex: runs where ChatGPT already is.
  9. Claude Code: built most of what SMGP runs on. Codex: built parts of our website.
  10. Which one first?
  11. First agent in the company? Start with Claude Code.
  12. Team lives in ChatGPT? Add Codex.
  13. Either way: install it, add an instruction file, give it one real task.
  14. Choosing the agent is the smallest decision.
  15. AI engineering is pointing it at the software your company runs on.
  16. SMGP. Embedded GTM engineering and embedded AI engineering. smgp.co

Why compare Claude Code and Codex at all?

If you have decided that AI should be building things inside your company, the first practical question is which coding agent to put on your machine. Two names come up every time: Claude Code, from Anthropic, and Codex, from OpenAI. They do the same kind of job and they are both improving quickly, which makes the choice feel bigger than it is.

We run SMGP on both, so this is not a review from the outside. It is a comparison from a company that uses each one every week, written for a business owner rather than a developer. The short version is at the top of this page. The rest explains it.

What is Claude Code?

Claude Code is Anthropic’s coding agent. It reads your codebase, edits files, runs commands and works with the tools around it, and it takes standing instructions from a file called CLAUDE.md in the project. It can read an AGENTS.md file too, if you already keep one.

It runs in more places than any agent we know of. There is the terminal, extensions for VS Code and JetBrains, a desktop app for Mac and Windows, and a version in the browser at claude.ai/code. It also plugs into Slack and GitHub Actions, can be picked up from your phone with Remote Control, and can run on a schedule with routines.

Around the agent sits a set of ways to shape it: hooks that run your own commands before or after an action, skills that package a repeatable workflow, subagents that split a job across several workers, and an Agent SDK for building your own agents on the same engine. You need a Claude Pro, Max, Team, Enterprise or Console account to use it.

What is Codex?

Codex is OpenAI’s coding agent. In OpenAI’s own words, Codex CLI is a coding agent that runs locally on your computer. You describe a task in the terminal, it reads the folder, edits the files and runs the commands, and it takes its standing instructions from an AGENTS.md file in the project.

Codex runs in the terminal, as an extension inside code editors, in the cloud at chatgpt.com/codex, and inside the ChatGPT desktop app. The desktop app, the terminal and the editor extension share one MCP configuration, so a tool connected once is connected everywhere. You sign in with a ChatGPT account, and OpenAI recommends a Plus, Pro, Business, Edu or Enterprise plan.

Claude Code vs Codex, side by side

What to compareClaude CodeCodex
Made byAnthropicOpenAI
The account you sign in withClaude Pro, Max, Team, Enterprise or a Console accountChatGPT Plus, Pro, Business, Edu or Enterprise, or an API key with extra setup
Where it runsTerminal, VS Code and JetBrains, a desktop app for Mac and Windows, the web at claude.ai/code, Slack, GitHub ActionsTerminal, an editor extension, the cloud at chatgpt.com/codex, the ChatGPT desktop app
The instruction file it readsCLAUDE.md, and AGENTS.md if that is what you haveAGENTS.md
Outside toolsMCP, added with claude mcp add and checked with /mcpMCP, added with codex mcp add and checked with /mcp
Where its settings liveA .claude folder and a .mcp.json file in the projectA config.toml file in your home folder
Ways to shape itHooks, skills, subagents, routines, Remote Control, the Agent SDKCheck OpenAI’s documentation for the current list
What it built at SMGPThe finance software, the company memory, most of the automationsParts of the smgp.co website

What do they have in common?

More than the marketing suggests. Both read a codebase, edit files and run commands. Both support MCP, the open standard for connecting an agent to outside tools, so either one can reach your automation platform, your CRM or your own data. Both keep a plain text instruction file in the project so you stop repeating yourself. Both run in a terminal and inside a code editor, and both come with a subscription your team may already have.

Both also pause for approval. When the agent wants to run a command or change something you have not already allowed, it asks. That is the feature that makes either one safe to hand to a founder who is not a developer, and it works the same way in each.

Where do they differ in practice?

Where you can use it. Claude Code is available on more surfaces today: the terminal, two editor families, a desktop app, the web, Slack and GitHub. If you want to start a task at your desk and check on it from your phone, or have an agent review every pull request in GitHub, Claude Code has that built in. Codex covers the terminal, the editor, the cloud and the ChatGPT desktop app.

What sits around it. Claude Code ships with hooks, skills, subagents, routines and an Agent SDK, which matters once you move from one-off tasks to systems that run on their own. We lean on those pieces heavily. Codex is moving fast too, so read OpenAI’s documentation for what it offers on the day you decide.

The account your team already has. This is the quiet deciding factor in most companies. A team that already lives in ChatGPT has the Codex sign-in ready. A team on Claude has Claude Code ready. Paying for a second subscription is reasonable, but it should be a choice, not an accident.

The instruction file. Claude Code reads CLAUDE.md and can read AGENTS.md. Codex reads AGENTS.md. If you ever want both on one project, write your rules in AGENTS.md once and both agents follow them.

Which one should you start with?

Start with Claude Code if this is your company’s first coding agent. That is the agent we built SMGP on, it runs on the most surfaces, and the pieces around it (hooks, skills, subagents, routines) are what turn a clever assistant into systems that run without you. The install is one command, and our guides for a Mac and a Windows PC walk through it.

Add Codex when you have a reason. Two good ones: you want a second opinion on a piece of code, which is genuinely useful, or your team already works in ChatGPT and the sign-in is already there. Our Codex install guide covers a Mac and Windows in one post.

Whichever you pick, do the same three things first. Install it, add an instruction file to the project, and give it one real task from your week. The tool matters less than the habit of handing work to it.

How does SMGP use both?

Claude Code built the finance software our company runs on, the company memory that files every call and decision where the next conversation can find it, and most of our automations. Those systems are described on how SMGP runs on AI, with the count we publish there: 455 automations built, 194 running today. Lucas, our founder, runs those sessions himself.

Codex built parts of this website. We use it when we want a second set of eyes on code or a different approach to a build, and we keep an AGENTS.md file in the repository so both agents work from the same rules. That split is not a verdict on either tool. It is where each one earned its place in our week.

Where does this fit in an AI-run company?

AI engineering is the job of pointing agents like these at the software a company runs on: invoicing, reporting, company memory and internal tools. Choosing between Claude Code and Codex is the first decision in that job, and it is the smallest one. The bigger decisions are what to build first and who owns the result, which is what embedded AI engineering is about.

If you want to see what a company looks like when that work has been done, how SMGP runs on AI lists the systems we run on, with dates. If you are still deciding whether the first hire should point AI at winning customers or at running the company, our post on the difference between a GTM engineer and an AI engineer is the place to start.

Lucas Mack

Written by Lucas Mack

Lucas is the founder and CEO of SMGP. He has done every job in the company, runs the strategy and the sales, and built every system SMGP runs on, from the outbound engine to the finance software. He writes about what he builds and what he has learned building it.

Share this post on LinkedIn (opens in a new tab)

Questions people ask

Can I use Claude Code and Codex on the same project?

Yes. Both work on ordinary folders of files, so nothing stops you running one after the other. Claude Code reads a CLAUDE.md file for its standing instructions and can also read an AGENTS.md file, which is the file Codex reads. Keeping your rules in AGENTS.md is the simplest way to share them between the two.

Do I need to be a programmer to use either one?

No. You need to open a terminal, paste one install command, and describe what you want in plain English. Both agents explain what they are about to do and ask before they do anything you have not allowed. The skill that matters is knowing your business well enough to describe the job.

Which one costs less?

Both bill through a subscription, and the plan your team already pays for usually decides it. Claude Code needs a Claude Pro, Max, Team, Enterprise or Console account. Codex is recommended with a ChatGPT Plus, Pro, Business, Edu or Enterprise plan. Limits change often, so check each company's plans page before you decide on cost alone.

Do both run on Windows?

Yes. Claude Code installs natively on Windows with one PowerShell command, and Codex has its own PowerShell installer. Both also run inside WSL, the Windows Subsystem for Linux, if your projects already live there. Our install guides cover each one step by step.

Do they both support MCP?

Yes. MCP is the open standard both agents use to reach outside tools such as an automation platform. In Claude Code you add a server with the claude mcp add command and check it with /mcp. In Codex you add one with codex mcp add and check it with the same /mcp command inside a session.

Talk it through with Lucas

Thirty minutes on your business: where a GTM engineer or an AI engineer would make the biggest difference, and what to build first. You leave with a plan either way.

Book a strategy call Contact our team