Claude is the model family from Anthropic; Claude Code is the command-line agent that runs it inside a repository. You type what you want, it reads the files it needs, edits them, runs the command, reads the error and tries again. Most of what is on this site started as a sentence typed into it.

How I use it

I work in the terminal, one session per project, and I treat it like a very fast junior developer who has read the whole codebase. The useful habit is to make it show its work: run the test, paste the output, re-read the file after the edit. When I skip that step, it occasionally tells me something is done when it is not, and that is the one trait you have to manage.

What surprised me is how much of the value sits outside code. It writes the commit message, drafts the German version of a page, reads a 2,000-line log and tells me which line matters. The second surprise was skills and subagents: a repeat task like "generate a cover image for this post" becomes a small file the agent reads before starting, and the next time I do not have to explain it again.

What annoys me: the subscription rate limits. On a long build day I hit the window and sit for an hour. The flat Max plan takes the invoice anxiety away, but not the wait. And every session starts from zero, so I keep notes in the repo for the next one. That is a discipline, not a feature.

Who it is for

Anyone who writes code and is willing to review what an agent produces. If you want a chat window with a code block in it, the free tier is enough. If you want the agent in your repo, the Pro plan is the right first step and Max is for people who use it all day. Skip it if you are not going to read the diffs; it needs a reviewer, and that reviewer is you.

Related tools