Anthropic's AI coding assistant, open-sourced March 2026

An AI that doesn't just
talk about your code —
it reads, writes, and runs it.

Claude Code lives in your terminal, understands your entire project, and makes changes with your permission. Think of it as a brilliant coworker who has already read every file.

~/my-project

$ claude

You: Why is the login test failing?

Reading src/auth/login.test.ts

Reading src/auth/login.ts

Found it. Line 42 expects the session to have a user field, but after your recent change to the expiry logic, expired sessions return null for user.

I can fix this by adding a null check before accessing user.id. Want me to make the change?

You: Yes, fix it

Editing src/auth/login.ts

Running npm test -- auth/login

All tests passing. Fixed and verified.

Like a senior developer who's
already read every file in your project

You describe what you need in plain English. Claude Code figures out which files to look at, what changes to make, and what commands to run — then does it, step by step, explaining as it goes.

chat STEP 1

You talk

Describe what you need in plain English. Ask a question, request a change, or describe a feature.

psychology STEP 2

It thinks

Reads relevant files, searches your codebase, understands the context. Figures out the right approach.

construction STEP 3

It acts

Writes code, edits files, runs commands, executes tests. Does the actual work, step by step.

verified_user STEP 4

You approve

Nothing changes until you say yes. Review every proposed change before it happens. Always in control.

This loop repeats as many times as needed — Claude Code can use multiple tools per question, working through complex tasks automatically.

Ask it anything. Watch it work.

Real examples of what you can ask Claude Code to do.

bug_report

"Why is this test failing?"

Reads the test, traces the logic, finds the bug, explains it in plain English, and offers to fix it.

Reads files → Analyzes → Explains → Fixes

add_circle

"Add dark mode to the settings page"

Finds your components, identifies the styling system, creates the toggle, writes the CSS, and tests it works.

Explores → Plans → Writes code → Verifies

school

"What does this codebase do?"

Scans the project structure, reads key files, and writes a clear summary. Perfect for onboarding to a new repo.

Scans structure → Reads files → Summarizes

transform

"Refactor auth to use JWT tokens"

Plans changes across multiple files, executes them step by step, then runs your existing tests to confirm nothing broke.

Plans → Edits multiple files → Runs tests

rocket_launch

"Create a PR for this feature"

Stages your changes, writes a meaningful commit message, pushes the branch, and creates a pull request with a description.

Stages → Commits → Pushes → Creates PR

groups

"Investigate and fix all the failing tests"

Splits the work into parallel tasks — like cloning itself to investigate multiple failures simultaneously, then fixing each one.

Splits work → Parallel investigation → Fixes all

You're always in control

Like a contractor who shows you the blueprint and asks "should I go ahead?" before making changes. Nothing happens without your approval.

PROPOSE

Claude Code wants to edit src/auth/login.ts

- if (session.user.id) {

+ if (session.user?.id) {

Green = being added. Red = being removed.

REVIEW

You see exactly what will change and decide:

check Yes, make this change

close No, don't change this

edit Try a different approach

SAFE

Only after you approve does anything change. Plus:

undo

Works with git, so every change is reversible

tune

Configure which actions need approval — start strict, loosen as you build trust

visibility

See everything it does — full transparency, no hidden actions

Start in under a minute

No configuration needed. Works with any codebase.

1

Install

$ npm install -g @anthropic-ai/claude-code

2

Go to your project

$ cd your-project

3

Start talking

$ claude

You: What does this project do?

It will guide you through a quick login on first run. After that, just type claude and start talking.

An AI that doesn't just
talk about your code —
it reads, writes, and runs it.

Like a senior developer who's
already read every file in your project

You talk

It thinks

It acts

You approve

Ask it anything. Watch it work.

You're always in control

Start in under a minute

Install

Go to your project

Start talking

That's just the beginning

How it thinks

Connect your tools

Work at scale

Under the hood

An AI that doesn't justtalk about your code —it reads, writes, and runs it.

Like a senior developer who'salready read every file in your project

You talk

It thinks

It acts

You approve

Ask it anything. Watch it work.

You're always in control

Start in under a minute

Install

Go to your project

Start talking

That's just the beginning

How it thinks

Connect your tools

Work at scale

Under the hood

An AI that doesn't just
talk about your code —
it reads, writes, and runs it.

Like a senior developer who's
already read every file in your project