The Data Practitioner
Chat answers. Agents act.
Claude Code and Codex, explained on one page for people who build Power BI reports rather than software.
The loop you already run
You ask Claude or ChatGPT for a DAX measure and it writes a good one. You copy it, click into Power BI, paste it, and check that it landed. When something breaks, you paste the error back and go again. The AI writes the code. You are the courier.
Claude Code and Codex remove the courier. Same request, same intelligence, but the AI opens your files, makes the edit, and checks its own work.
The chat window
Claude · ChatGPT
New Customer Revenue =
CALCULATE ( [Total Revenue], ... )Then it is on you: copy, switch to Power BI, paste, test. Every error makes the round trip again.
The agent
Claude Code · Codex
Then it is on it: you read the change and approve it. No copy, no paste, no round trips.
Who is who
Each company makes a chat product that answers you and an agent that acts for you. They run the same underlying AI models; the difference is not intelligence, it is hands.
| Chat: answers you | Agent: acts for you | |
|---|---|---|
| Anthropic | ClaudeThe website and apps at claude.ai. You talk, it answers, you carry the results. | Claude CodeRuns in your terminal, including the one inside VS Code, and as a desktop app. Reads your files, edits them, runs commands. |
| OpenAI | ChatGPTThe website and apps at chatgpt.com. Same deal: great answers, your hands. | CodexOpenAI's agent, same idea as Claude Code: terminal or desktop app. Everything on this page applies to both. |
Why this works on Power BI
Agents edit text files, and a Power BI project can be text files. Save your work as a PBIP project and the semantic model and report definition become folders of plain, readable text. That is exactly the shape these agents were built to work on. The same tool developers point at software projects works on your measures, your tables, and your report pages.
Three worries, handled
Is the terminal a developer thing?
A terminal is a text box: you type a command, press Enter, and read the reply. Once an agent like Claude Code is running inside it, though, what you type is plain English. VS Code ships with one built in, and both agents live comfortably inside it.
Will it change my files without asking?
No. Before it runs a command or edits a file, you get a yes-or-no prompt, and every edit is reviewable before you keep it.
Do I need another subscription?
Probably not. Claude paid plans include Claude Code, and ChatGPT paid plans include Codex. If you already pay for the chat, you likely already have the agent.
Start here
- Install Claude CodeOfficial docs and the one-line install.
- Meet CodexOpenAI's overview; the CLI itself is on GitHub.
- Save your work as a PBIP projectThe step that turns your report into files an agent can edit.
- Run one small sessionOpen VS Code on a copy of a project, open the terminal, and ask for one small change.