What Claude Code is (and what it isn't)

Before you install Claude Code, it's worth a sharp picture of what it actually is, because the name doesn't quite give it away. The clearer your picture, the faster you'll get value out of it.

The one-sentence version

Claude Code is an AI agent that lives in your terminal, reads and edits files on your computer, runs commands on your behalf, and asks before doing anything that could matter.

Read that sentence twice. Every phrase matters.

• AI agent— it doesn't just generate text, it takes actions.
• Lives in your terminal — not a website, not a browser tab. It runs as a program on your machine.
• Reads and edits files — your actual files, not copies. The same ones VS Code is showing you.
• Runs commands on your behalf— git, tests, installs, anything you'd type yourself.
• Asks before doing anything that could matter — there are guardrails. You stay in the loop.

What it isn't

Equally important — three things people often think Claude Code is, that it isn't:

It isn't a chatbot

Chatbots generate text and stop. Claude Code can act: open a file, change a line, run a test, see what broke, change another line. The conversation is the steering wheel, not the product.

It isn't autocomplete

Tools like GitHub Copilot live inside your editor and suggest the next few characters as you type. They're great for that. Claude Code is doing something much bigger: planning multi-step changes, navigating a whole codebase, running and verifying its own work. It operates at the level of a task, not a keystroke.

It isn't a code generator you copy from

Plenty of people use AI as a fancy answer engine — ask a question, get a code snippet, copy it into the editor. That works, badly. The snippet doesn't know about the rest of your project, your naming conventions, your existing utilities, your tests. Claude Code does, because it actually opens those files first.

The agent loop

Internally, every interaction with Claude Code follows the same loop:

1. Read your message. What do you want?
2. Think. What does the agent need to know to do this well? What files should it look at first?
3. Use a tool. Read a file. Search for a pattern. Run a command. Edit a file. Each is a discrete action.
4. Observe the result. Did that work? What did the file actually say? What was the test output?
5. Loop. Decide what to do next based on what just happened, until the task is done.
6. Report back.Tell you what it did, what it learned, what's next.

This loop runs for every request. Sometimes it's a single cycle (you ask a simple question; Claude reads one file and answers). Sometimes it runs dozens of times in a single response (you ask for a complex feature; Claude reads, edits, tests, edits again).

The three workflows it's built for

Most of what you'll do with Claude Code falls into one of three shapes:

• Greenfield.Empty folder, blank canvas. "Build me a website that does X." Claude scaffolds everything from scratch.
• Maintenance.An existing project. "Add a dark mode toggle." "Fix the bug where the login form submits twice." Claude reads the existing code first, then edits in a way that fits.
• Exploration.Code you didn't write. "Explain what this repository does." "Where is the function that handles refunds?" Claude as a reading companion.

All three use the same agent loop. The difference is mostly how much existing context Claude has to load first.

The shift it asks of you

Using Claude Code well isn't about typing faster or memorising prompts. It's about shifting your role: from typing code yourself to describing intent, reading results, and steering.

That's a real change. Some of your existing instincts will help (decomposing a problem, judging whether output is correct). Some will hurt (the urge to micromanage every line). The rest of this module is about building the right habits.