Claude Code and Codex come up in almost every message in the group. They were built for programmers and their names sound technical, but they do much more: they organise documents, pull data out of invoices, prepare reports, build a website or automate a process at work. Coding is just one of the things they use to get there, and you don’t need to code to ask for it.

What Claude Code and Codex are

Claude Code is Anthropic’s agent. It came out in February 2025 as a program for the terminal, and today it runs in the Claude desktop app for Mac and Windows, in the terminal, on the web, in code editors and even from your phone. Since 22 September 2026 it uses Claude Opus 5.5 by default.

Codex is OpenAI’s agent. Its big advantage is that it comes with ChatGPT. Since July 2026 it lives inside the ChatGPT desktop app (it used to have its own), and it also runs on the web, in the terminal and in code editors. On paid plans it uses the GPT-6 family: it starts with Sol and keeps Astra for the hardest jobs; on the free account you get Luna.

If agents still sound strange to you, start with what an AI agent is and come back.

Is Claude Code free? What they cost

Claude Code isn’t free: it comes with Claude Pro (in Spanish), which costs €18 a month plus VAT, about €22. Codex comes with ChatGPT, and Plus expands its usage; the reference checked on 24 September was about €23 a month including VAT. If you just want to try an agent without paying, Codex works with a free ChatGPT account, with limits that depend on the task.

A fee of about €20 does not buy a fixed number of tasks. The model, files, tools and amount of work each task needs affect usage. Check limits and reset windows in your account; credits, extra usage and the API can cost more. Claude Max offers 5 or 20 times the usage of Pro, but this does not guarantee working hours or a return. In which AI subscription is worth it I explain how to compare your needs with the full cost.

Claude Code or Codex: which one to choose

If you already pay for ChatGPT, start with Codex. If you pay for Claude, or you’re about to pick a plan, go with Claude Code (it needs Pro). And if you only want to try one for free, Codex on a free ChatGPT account. They work in a very similar way and what you learn with one works with the other. Right now mine is Claude Code, above all for long jobs, and if that changes I say so in the group.

How to use Claude Code without the terminal

  1. Download the Claude app from the official website, for Mac or Windows.
  2. Sign in with the account that has the Pro plan or higher.
  3. Go to the Code section. Depending on your account you will see two tabs, Chat and Code, or three, with Cowork in the middle. The agent lives in Code.
  4. Pick a test folder. The agent will work inside the folder you point it to.
  5. Choose Manual where available in the permission selector and review existing permissions. Auto uses automated review; its decisions can be wrong.

Codex is similar: download the ChatGPT app, sign in and open your test folder in Codex. Review its permissions: edits and local commands may already be approved without another prompt.

Start with fictional documents in a test folder and keep the originals separately. A copy protects the original from loss; it does not prevent leaks or limit agent access by itself.

What to answer when it asks for permission

Manual does not ask about everything: some reads and actions are already allowed. Check the settings and read each request. If you do not understand a command, ask what it will do before approving it. Expand permissions only when the task needs it and you can review the result.

Five tasks for your first hour

  1. Have it explain the folder. It only reads, nothing changes.
  2. Organise it, plan first. It shows you the plan and waits for your go-ahead.
  3. Pull data into a spreadsheet. The task that saves the most time.
  4. A report from that data. One page with a chart, to show a partner.
  5. A small custom tool. For example, a quote calculator that runs on your computer.

The first one, to copy as is:

Look at all the files in this folder and tell me what is here: how many files, what type, from which dates and what they are about. Don't change or move anything.

When it mentions something you don’t follow (an MCP, a skill, an AGENTS.md), it’s all explained in plain words in the AI agent dictionary.

What about Claude Cowork? How it differs

Cowork is Anthropic’s agent for office work, built for people who don’t want to see anything technical: you give it a folder or your apps and it hands back documents, spreadsheets or reports. Since September 2026 Anthropic has been moving it into the normal Claude chat. OpenAI’s equivalent is ChatGPT Work.

If your work is mostly documents and email, that may be all you need. I compare them in Claude Cowork vs ChatGPT Work. If you want more (automate a process, build a tool, work through hundreds of files), Claude Code and Codex go further.

When it says it’s finished

Ask it to prove it: «Check that it works and show me the proof». That makes it verify its own work. Then review it yourself, the way you would review the work of someone who just started with you. And since it works with copies in your test folder, nothing is lost if something goes wrong.

Review on 30 September 2026: usage limits and buying criteria. Price references retain their stated date; confirm the current total for your account before paying.

Sources, checked on 26 September 2026: Claude prices (in Spanish), Claude Code overview, its desktop app, Claude Code models, permission modes, the Claude Code launch, Codex models, Codex by plan, Codex permissions and ChatGPT release notes.