What is Claude Code?

Claude Code is an agentic AI coding assistant that runs in your terminal. If that sentence means nothing to you yet — perfect. This page unpacks every word of it, assuming zero background.

Breaking Down the Words

”AI coding assistant”

A program powered by a large language model (Claude, made by Anthropic) that helps you write software. You talk to it in normal English — “fix the login bug”, “add a dark mode toggle” — and it understands both your words and your code.

You may have used AI chat tools that suggest code which you then copy-paste. Claude Code is a different category, and the difference is the next word.

”Agentic” — the word that changes everything

An agent is an AI that doesn’t just talk — it acts. Claude Code has hands:

A chat AI can…	An agentic AI can ALSO…
Explain what code does	Open and read your actual files itself
Suggest a fix as text	Edit the file and apply the fix
Tell you a test command	Run the tests and read the results
Describe a git workflow	Make the commit and open the PR
Answer from training data	Search the web / your docs when unsure

And crucially: it checks its own work. If the tests fail after its edit, it reads the failure and fixes it — without you doing anything.

”Runs in your terminal”

The terminal (command line) is the text window developers use to control their computer (PowerShell on Windows, Terminal on Mac). You type claude there, and the assistant starts inside your project folder — which is why it can see and change your real files. (It also runs in VS Code, a desktop app, and the browser — same engine everywhere.)

The Agentic Loop — How It Actually Works

Everything Claude Code does is this cycle, repeating until the job is done:

        ┌────────────────────────────────────────────┐
        │  1. GATHER CONTEXT                          │
        │     read files, search code, run commands   │
        ▼                                             │
   2. TAKE ACTION                                     │
      edit code, create files, run tools              │
        │                                             │
        ▼                                             │
   3. VERIFY RESULTS                                  │
      run tests, check output ──── not done yet? ─────┘
        │
        ▼ done ✓
      report back to you

You can interrupt at any time with Esc to redirect it.

A Real First Conversation (what it actually looks like)

you>    there's a bug where users can submit empty forms - fix it

claude> I'll investigate the form handling.
        ● Read(src/components/ContactForm.tsx)
        ● Grep("onSubmit", src/)
        Found it — handleSubmit() never validates fields before posting.
        ● Edit(src/components/ContactForm.tsx)   [shows you a diff]

you>    [press Enter to approve the change]

claude> ● Bash(npm test)
        All 24 tests pass, including 2 new ones I added for empty-field
        validation. The bug is fixed.

Notice what you did NOT do: open a file, write code, or run a test. You described an outcome and approved a change. That’s the skill this whole curriculum teaches — directing an agent instead of typing every line yourself.

What It Can Do (capability map)

Understand entire codebases across files and languages
Edit files, write tests, create commits, open pull requests
Run any shell command — builds, tests, deploys
Search the web and fetch documentation
Connect to your other tools via MCP: GitHub, databases, Slack… (25)
Remember your project’s conventions across sessions (17)
Delegate to helper agents and work in parallel (28+)
Run scheduled/automated jobs while you sleep (34)

Is It Safe to Let an AI Touch My Files?

By default, Claude Code asks your permission before every file change and command. You see a diff, you approve or reject. As you gain trust you can loosen this gradually (auto-accept edits → sandboxed commands → full autonomy in containers) — the curriculum covers exactly how and when: 14, 16. And /rewind can undo its changes anytime (08).

Where It Runs

Platform	How
Terminal	`claude` CLI — the main surface, what this course teaches
VS Code	Extension
JetBrains	Plugin
Desktop	Standalone app
Web	claude.ai/code
CI/CD	GitHub Actions, GitLab
Slack	`@Claude` mentions → cloud sessions
Chrome	Browser automation via extension
Mobile	Claude iOS app + Remote Control

Core Models (the “brains” you can choose)

Alias	Resolves to	Best for
`fable`	Claude Fable 5	Hardest, longest-running autonomous tasks
`opus`	Claude Opus 4.8	Complex reasoning
`sonnet`	Claude Sonnet 5	Daily coding (default)
`haiku`	Claude Haiku 4.5	Fast, lightweight tasks
`opusplan`	Opus + Sonnet	Opus plans, Sonnet executes

Switch anytime with /model <alias>. Full details: 13-model-config-env-vars.md

Words You’ll Keep Hearing

New terms are explained as they appear, but three you need immediately:

Token — the unit AI text is measured in (~¾ of an English word). Models have limits and pricing per token.
Context window — Claude’s working memory for one session. Big but finite; managing it is a core skill (09).
Prompt — whatever you type to the AI. Better prompts → better results (02 has patterns).

Full A–Z reference: 57-glossary.md

🧠 Quick Recall

Trick: “An agent is AI with hands.”
The loop = GAV: Gather → Act → Verify.
Chat suggests; an agent does — and checks its own work.

👨‍🏫 Teach It

Say: “You describe outcomes; it does the typing and the testing.”
Show: fix a real bug in a real repo in 3 minutes.
They’ll trip on: thinking it runs wild — by default it asks before every change.

Learning path: Index · ➡ 01-installation-setup.md

Written by Fenil Patel