Claude Code vs Codex: Setup, Limits, and Using Both

Have you spent hours reading Claude Code vs Codex comparisons, still unsure which AI coding agent to start with? Pick one without trying both and you may only find out it doesn't fit your work after you've installed it and written all your instructions.

Claude Code is Anthropic's coding agent, and Codex is OpenAI's. Both read the folders on your computer, edit files, and run commands, and both let you shape how they work with instruction files, skills, and MCP. This guide compares installation, sign-in, instruction files, skills, plugins, MCP, and usage limits based on both official docs as of November 2026. It also shows how to test the two on the same task and how to run them side by side. At the end, you'll see how to turn a meeting transcribed with daglo into a task list your agent can work from.

Claude Code vs Codex at a glance

The two tools can do almost the same jobs. The differences are in accounts, instruction file names, and configuration. If you already pay for Claude, start with Claude Code. If you're on a ChatGPT plan, start with Codex.

Both handle long tasks, multi-file changes, and code review. So instead of asking which one is smarter, ask which one fits the accounts and tools your team already uses.

Install, sign in, and set up instruction files

You install both with one terminal command and sign in the first time you run them in a project folder. Run /init right after installing, and the agent reads your project and drafts an instruction file.

1. Install Claude Code: On macOS or Linux, run curl -fsSL https://claude.ai/install.sh | bash. In Windows PowerShell, run irm https://claude.ai/install.ps1 | iex. Open a new terminal and check with claude --version.
2. Install Codex: On macOS or Linux, run curl -fsSL https://chatgpt.com/codex/install.sh | sh. You can also use npm (npm i -g @openai/codex) or Homebrew.
3. Sign in: Run claude or codex in your project folder and follow the prompt. For Codex, choose "Sign in with ChatGPT."
4. Write instructions: Add your build and test commands, files the agent must not touch, and what "done" means to the draft /init created.

Instruction files load into every conversation and use up your limits, so keep them short and focus on rules the agent gets wrong. For the bigger picture of designing instructions, tools, and permissions together, see our harness engineering guide.

Skills, plugins, MCP, and usage limits

A skill is a workflow the agent loads only when needed. MCP is the standard that connects an agent to outside tools such as Notion or Slack. Both tools support the same SKILL.md skill format and MCP, so a workflow you build once works in either.

Limits and plans change often, so check the official pricing pages before you start. For details, read our guides to Claude Code skills and setting up Codex with AGENTS.md.

How to compare them on the same task

The surest way to choose is to give both tools the exact same request. Here, you'll build a script that cleans up a downloads folder. Run it on a copied test folder, not your real one.

Paste this request into both tools:

Write a Python script that moves files in the test-downloads folder into Images, Documents, Archives, and Other folders by file extension. Add a preview mode that only lists what will move, and never overwrite files with the same name.

Score each result on five points:

• Plan: Did it explain its approach before writing code?
• Approval: Did it ask before moving files?
• Preview: Did the preview mode actually make it in?
• Verification: Did it run the script and check the result itself?
• Usage: How much did /usage or /status show afterward?

Make the one that worked more the way you want your default. In Claude Code, press Shift+Tab for plan mode to see a plan first. In Codex, set the approval policy to on-request so it checks with you when needed.

Use both together, and hand off meeting notes as context

To run both, keep one AGENTS.md as the single source of instructions. Claude Code can read AGENTS.md, so add one line, @AGENTS.md, to CLAUDE.md. Then editing one file keeps both agents on the same rules.

Most of what you hand an agent gets decided in meetings. Transcribe the meeting with daglo and you get a record of who said what, ready to use as context. daglo's Speaker ID remembers the voices of the people in your meetings and fills in their real names instead of "Speaker 1" from the next recording on. It's the first feature of its kind in Korea offered to every user, so you no longer fix names by hand.

1. Record the meeting, transcribe it with daglo, and ask Board Chat: "List only the decisions and action items."
2. Copy the answer into meeting-notes.md in your project folder.
3. Ask Claude Code or Codex: "Read meeting-notes.md and turn the action items into a checklist in tasks.md."

If you run the same summary after every meeting, save that request in daglo's My Templates. daglo remembers the prompt and summarizes every new recording in that format automatically.

In short: pick the tool that matches the subscription and team tools you already have, compare the two on one small task, and share a single AGENTS.md if you use both. Commands and plans change often, so check both official docs once more before you set things up.

#ClaudeCode #Codex #AICodingAgents #AGENTSmd #Automation #AIAgents #DeveloperTools #daglo

👉 Transcribe your next meeting free