Claude Code, the autonomous dev agent that runs on your machine and touches your files.
This chapter is the entry pitch. The dedicated guide has 19 chapters and covers install, workflows, sub-agents, MCPs, and hooks. Head to the Claude Code guide when you want to go deep.
Who it's for
Developers who want an AI teammate plugged straight into their codebase. Technical PMs who want to read code without bothering the team. Founders building their own software who don't have ten hours a day to spend in an editor.
And also non-devs. You describe what you want in plain English, Claude Code writes the code, tests it, runs it. Prototyping an app, automating a local script, hacking on a site, all of that becomes reachable without a year of training.
The 5 key capabilities
Read and write your files directly. Claude sees your project, opens the right files, changes what needs changing. You stay in control of the diffs before you approve.
Run commands in your terminal. Tests, build, database migrations, deploy. It executes, you watch the output.
Plan then execute full features. Plan mode: it proposes a strategy, you approve, it goes. No surprises at the end.
Delegate to sub-agents in parallel. One agent explores the code, another writes tests, a third checks security. You orchestrate.
Hook up MCPs to act on external services. Notion, Vercel, Supabase, Figma, Gmail. Claude is no longer stuck on your machine, it reaches into your tools too.
How much it costs
Included in the Pro plan and above on claude.ai. Pro at $20/month already gives you enough volume for serious work. Max at $100 or $200/month unlocks much bigger quotas if you're coding with it all day.
Limits are expressed in tokens per 5-hour window, not in number of messages. You feel the plan difference when you stack up big refactors.
When to use it vs something else
Claude Code when you want to act on code, files, local commands. Anything that touches your disk or your git repo goes here.
Claude.ai (the chat) when you want to think, write, brainstorm an architecture, prep a spec. No executing agent needed, just a brain to talk to.
The full Claude Code guide is here, 19 chapters plus a terminal appendix. See the guide.