Installing Claude Code takes about a minute. Knowing what to do in the first hour is what decides whether it sticks. Here’s both.
Install
The native installer works on macOS, Linux, and Windows (WSL):
curl -fsSL https://claude.ai/install.sh | bashOr, if you live in the Node ecosystem:
npm install -g @anthropic-ai/claude-codeThen cd into a real project — not an empty folder, it needs code to be useful — and run:
claudeFirst run walks you through signing in with your Claude account (Pro or Max subscription) or an Anthropic API key. If you’re not sure which, the pricing guide covers the trade-off in two minutes.
Your first hour, in four moves
1. Make it explain before it edits. Ask “how does auth work in this codebase?” and watch it read files and assemble an answer. This builds the right mental model: it’s an agent that explores, not a chatbot that guesses.
2. Give it a small, real task. A rename, a missing test, a TODO you’ve been avoiding. It will propose edits and ask permission before touching files or running commands — approve, and check the diff with git diff like you would any junior’s PR.
3. Create a CLAUDE.md. Run /init and Claude Code writes a project memory file — build commands, conventions, gotchas — that it reads at the start of every future session. Two minutes now, permanently better output.
4. Learn the one flag that changes everything. claude -p "your task" runs headless: no interactive session, just do the thing and print the result. This is the doorway from “AI pair programmer” to “agent I can schedule, script, and dispatch.”
Where people go next
Once the basics click, the pattern that compounds is delegation: keeping a list of scoped tasks and handing them to Claude Code instead of doing them interactively. Connect it to your to-do list over MCP and “assign this task to Claude” becomes a thing your task manager can just do — Lume was built around exactly that loop. And when you’re ready to teach it your team’s workflows, read up on skills.