Your first 30 minutes with Claude Code

If what Claude Code is made sense in theory but you haven’t actually opened it, here’s the first half hour, literally.

Minutes 0–5: open it. If you’re on Claude Desktop, click the Code tab — nothing to install. If you want the standalone CLI instead, run npm install -g @anthropic-ai/claude-code, then type claude in a terminal. Either way, point it at a real folder: a project you’re working on, or just a new empty one.

Minutes 5–10: ask it something real, not a test question. Don’t ask “what can you do.” Take an actual document you have — a messy notes file, a draft, a spreadsheet — and ask it to do something with that specific file: summarize it, find an error, draft a rewrite. Watch what happens to the actual file, not just the chat window.

Minutes 10–20: notice the diff. If it edited something, it should show you a before/after — a diff — rather than just handing you new text to paste in yourself. This is the whole point of Code over chat: it changes things in place, and you approve or reject the change. Reject one on purpose, just to see what that looks like.

Minutes 20–25: write a one-paragraph CLAUDE.md. In that same folder, create a file called CLAUDE.md with three or four lines: what this project is, what you’re trying to do with it, anything it should know before helping. Start a new session and notice it already knows this — that’s memory, and it’s the single highest-leverage thing you’ll set up all week.

Minutes 25–30: stop and decide if it’s worth a second session. You don’t need to learn everything today. If the diff-review workflow felt useful, that’s the signal to come back tomorrow with a slightly bigger task.

That’s the whole on-ramp. Everything else — slash commands, subagents, hooks, permissions — is depth you add once this feels normal, not prerequisites you need first.