What Is Claude Code and How It Differs from ChatGPT

Claude Code is Anthropic's command-line AI coding agent. Unlike a chat interface, it runs directly in your terminal and has direct access to your project — it reads your files, runs your tests, makes edits, and commits code without any copy-paste between you and a browser window. For QA engineers, this distinction changes everything about how AI fits into daily work.

Not another chatbot

The most common first question is "how is this different from Claude.ai or ChatGPT?" The answer is agency. Chat-based AI tools are conversation partners — you describe what you want, they respond with text, and you manually carry the output back to your project. Claude Code operates directly on your file system. It reads your actual test files, understands your project's patterns, and makes changes in place.

The second question is "how is this different from GitHub Copilot?" Copilot sits inside an IDE and suggests the next line or block as you type. It is an autocomplete tool. Claude Code accepts high-level, multi-step instructions: "look at how my login tests are structured, then write a similar test for the signup flow and run it to verify it passes." That is not something an inline autocomplete tool can do.

Cursor is an AI-native IDE that is genuinely powerful — but it is an IDE. Claude Code is a terminal tool that complements any IDE or editor you already use. They are not competing for the same workflow.

What Claude Code actually does in a QA session

A typical session might look like this:

> Look at tests/login/ and understand my test patterns. Then generate a Playwright test 
  for the signup flow at https://staging.myapp.com/signup. Run it when done.

Claude Code will read the files in tests/login/ to learn your conventions, navigate to the signup URL if Playwright MCP is configured (covered in Chapter 4), generate the test file and write it to disk, run the test and report the result, and fix straightforward failures — all without you leaving the terminal.

Key capabilities for QA

• Codebase-aware generation — reads your existing tests before writing new ones, so output matches your style and uses your existing helpers
• Bulk refactoring — find all tests using a deprecated API and update them in one pass
• Failure diagnosis — paste a stack trace or point at a failed CI run; Claude Code analyses it and suggests fixes
• Git integration — git diff, git commit, git log are available and Claude Code uses them naturally
• MCP server support — connects tools like Playwright MCP for live browser interaction during test generation

Why QA engineers reach for it

The pattern that matters most for testers is the closed feedback loop. You describe a task, Claude Code acts on it, you see the result, you iterate. No copy-paste. No context-switching. No manually threading an AI response back into a file.

For the kinds of work that eat QA time — writing boilerplate tests, refactoring selectors after a redesign, diagnosing an intermittent CI failure — that loop is the productivity multiplier. The AI Tools for QA course covers the broader landscape of AI tooling; this course goes deep on Claude Code specifically.

⚠️ Common Mistakes

• Treating generated code as final. Claude Code produces a first draft. Always read what it wrote before running or committing. Generated tests can pass while asserting the wrong thing.
• Using it like a chatbot. Asking "how do I write a Playwright test" gets you an explanation. Asking "look at tests/auth/ and generate a similar test for password reset at /reset" gets you working code.
• Skipping CLAUDE.md. Without project context in a CLAUDE.md file, Claude Code guesses at your conventions. Lesson 4 covers this file — it will save hours of repeated context-setting across sessions.

🎯 Practice Task

Before installing anything, map the friction points in your current AI workflow. 10 minutes.

1. Think of the last three times you used a chat-based AI tool (ChatGPT, Claude.ai) for a testing task.
2. For each one, note: how many times did you copy something between the browser and your editor? How many clarifying back-and-forths did it take?
3. Write one sentence describing the task you'd most want to hand off to an AI that could read your codebase directly.

Keep that task in mind — by the end of Chapter 2 you will have run it.

The next lesson walks through installing Claude Code and getting authenticated in under 10 minutes.