Setting Up Claude Code

I want to start with something the lesson almost says but doesn't quite commit to. There's a meaningful difference between using Claude in a browser and having it actually sitting inside your file system. That's not a minor UX upgrade — that's a different relationship with the tool entirely.

It is the difference between asking someone to help you move house over text message and having them actually show up. One of these is considerably more useful, and yet people spend months in the text message phase as though that is a reasonable way to live.

Though to be fair, the setup here isn't trivial. You're configuring environment variables, verifying file system permissions — there are legitimate ways it can go wrong before you've done anything interesting.

Legitimate, yes. Insurmountable, no. The lesson covers it. The instructions exist. What some people will do is encounter one error message and conclude that the whole enterprise is broken, when in fact they have mistyped their API key.

That's a specific prediction.

It is also a specific and recurring human behaviour.

What I think is genuinely underappreciated here is the mental model shift the lesson mentions — Claude as a local CLI agent rather than a chat interface. Most people's instinct is still to copy-paste. File open in the IDE, answer in the browser, back again. It's like insisting on faxing something you could just email.

The copy-paste workflow is, at its core, a person doing the integration work that the tool should be doing. You are the API. You are the bottleneck. That is a strange position to voluntarily occupy when there is an alternative.

Although I'd push back slightly on how the lesson frames the objective. 'Remove manual friction' is fine as far as it goes, but it undersells it. You're not just saving clicks. You're enabling a different kind of iteration — the model can read context you wouldn't have thought to paste, catch things you didn't know to ask about.

That is a fair point, and I will not pretend otherwise. The lesson treats this as a setup task, which it is, but the actual unlock is that Claude stops operating on the slice of your codebase you remembered to share and starts operating on the actual thing.

Which also means you should probably understand what it has access to. The file system permissions step isn't bureaucratic housekeeping — it's you deciding the scope of what you're handing over.

Most people will click through that part. They will then be surprised later, which is a surprise they have, in a technical sense, engineered for themselves.

You're not wrong, but that applies to almost every permissions dialogue ever written. The solution isn't to blame the person — it's to understand why that step actually matters, which the lesson does explain if you're reading carefully.

'If you're reading carefully' is doing considerable heavy lifting in that sentence.

Fine. Read the permissions section carefully. It tells you what Claude can touch. That's not a small thing to know before you type your first command.