Chapter 01 · Section I · 12 min read
What Claude Code actually is
A quick, honest description of Claude Code — what it does, what it does not do, and why "an agent in your terminal" is a more useful mental model than "a chatbot for code".
Claude Code is a command-line tool that turns your terminal into a working session with a large language model. The model can read your files, write new ones, edit existing ones, run shell commands, search the web, and coordinate all of that on your behalf — with you approving or vetoing what it wants to do. It is not another chat window. It is a coding assistant that acts on your workspace.
The distinction matters. If you have used claude.ai in a browser, you have used the model in “advisor” mode: you paste code, it replies with a suggestion, you paste the suggestion back. Claude Code closes that loop. When it decides a file needs a change, it proposes an edit; you approve it; the file changes. When it needs to know if the tests still pass, it proposes to run npm test; you approve; the output comes back into the conversation. The word for this in current AI parlance is agentic — the assistant does not just produce text, it takes actions.
What it is good at
- Reading a codebase. Give it a repo it has never seen and it can navigate the file tree, grep for symbols, and answer questions about how things fit together.
- Doing focused edits. “In
src/api/user.ts, add input validation tocreateUserthat rejects empty emails and returns a 400.” That kind of well-scoped ask is exactly the shape it wants. - Running a workflow end-to-end. Write the code, run the tests, read the failure, fix, rerun. It can do all four steps in one loop when you let it.
- Explaining unfamiliar code. It is honestly a very good tutor for parts of a codebase you inherited.
What it is not
- A replacement for judgement. It will happily do the wrong thing very quickly if you tell it to.
- A production system on autopilot. Everything it does is either your approval or a preconfigured allowlist. If either of those is set too loose, that is your problem, not the model’s.
- A silent tool. It writes back a lot of text. Reading that text is part of using it well.
Who this course is for
Software engineers who already ship code and want an honest map of what this tool can and cannot do. If you are new to programming, this is not the right starting point — start with Building AI in the tracks section first, then come back.
Check your understanding
Quick check
—