This chapter explains what Claude Code is, how it differs from other AI tools, and how the roles are split between you and it. Reading time: 5 minutes.
The idea in one sentence
Claude Code is an autonomous development agent that runs on your computer. You give it a direction, it writes all the code, creates the files, runs the checks, and fixes its own mistakes.
How is it different from ChatGPT or the Claude chat?
When you use ChatGPT or the chat on claude.ai, you copy-paste code between the conversation and your project. It's slow, error-prone, and you lose context with every message.
Claude Code runs directly on your computer. It sees all your files, understands the full architecture of your project, and acts on it in real time. It's the difference between asking for directions over the phone and having a copilot sitting next to you in the car.
| Criterion | Classic AI chat (ChatGPT, claude.ai) | Claude Code |
|---|---|---|
| File access | None, you copy-paste | Reads and writes directly |
| Command execution | No, you do it yourself | Yes, runs tests, the build, installs packages |
| Project context | Limited to what you paste | Sees the entire source code |
| Continuity across sessions | Starts from scratch each conversation | Automatically reloads the project's conventions |
| Self-correction | No, you check manually | Yes, detects and fixes its errors in a loop |
The 5 concrete superpowers
1. Read and write files
Claude Code navigates your project like a developer. It opens files, understands imports, follows dependencies. When it creates a component, it already knows what exists and integrates naturally.
2. Run commands
It launches the app, the tests, type checks, database migrations. When something fails, it reads the error message and fixes it before moving on.
3. Understand the global context
It doesn't work on a single isolated file. It understands the full architecture: how the pages call server actions, which call the database, which return typed values. When you modify a table, it knows which files need to be updated.
4. Plan and execute plans
You give it a multi-step plan, and it executes it task by task with checkpoints between each step. If a step breaks something, it stops and fixes before continuing.
5. Maintain consistency
Thanks to the CLAUDE.md file (which we'll cover in chapter 3), Claude Code respects your conventions across the entire project: same names, same patterns, same colors, same language.
The analogy: you = project manager, Claude = senior developer
Imagine you hire an ultra-fast senior developer, available 24/7, who never gets tired. He's technically excellent, but he needs three things from you:
You, the project manager - You describe what you want to build and why - You choose between the options Claude proposes (it always leaves you the choice) - You visually test the app in your browser (Claude doesn't see the screen) - You validate saves (commits) and deployments to production - You bring your business knowledge, you know what your users need
Claude Code, the developer - It proposes the technical architecture and design choices - It writes 100% of the code, not a single line from you - It runs automatic checks (types, lint, tests) and fixes them in a loop - It manages Git (the history of changes) with clear commit messages - It documents what it does and why
The 3 work modes
Conversation mode
You chat with Claude like a coworker. You ask questions, ask for explanations, explore ideas. Nothing is modified.
"How does the authentication system work in the project?"
"What options do I have to add a notifications system?"
Direct execution mode
You ask for a precise action, Claude executes it immediately.
"Add a 'Duplicate' button on the content card, which creates a copy as a draft."
"The signup form shows a 500 error when the email already exists. Fix it."
Planning mode
For complex features, you brainstorm first, plan, then execute the plan. It's the most powerful mode, we cover it in detail in chapter 5.
"Plan the implementation of a publishing streak system."
Key takeaway: Claude Code isn't a chatbot that hands you code to copy-paste. It's an autonomous agent that acts directly on your project. Your role is the vision and the decisions. Its role is the technical execution.