How AI coding tools work — Code with AI
Three generations in five years
Autocomplete — the AI finishes the line you're typing.
Chat — you paste code into a chatbot and copy the answer back.
Agents — the AI works inside your project: it reads files, makes edits across many of them, runs your app and tests, reads the errors, and fixes them — in a loop — until the task is done.
Today's tools — Cursor, Claude Code, Codex, Gemini CLI — are agents. You describe the outcome; they do the typing.
📝 Read the full lesson notes
Three generations in five years
Autocomplete — the AI finishes the line you're typing.
Chat — you paste code into a chatbot and copy the answer back.
Agents — the AI works inside your project: it reads files, makes edits across many of them, runs your app and tests, reads the errors, and fixes them — in a loop — until the task is done.
Today's tools — Cursor, Claude Code, Codex, Gemini CLI — are agents. You describe the outcome; they do the typing.
Editor vs. terminal vs. cloud
In an editor (IDE): Cursor is a full code editor with AI built into everything. Claude Code and Codex also ship extensions for VS Code and friends.
In the terminal: Claude Code, Codex CLI and Gemini CLI run as command-line agents — powerful, scriptable, editor-agnostic.
In the cloud: hand off a task from a browser or phone; the agent works in a remote sandbox and comes back with changes to review.
"Vibe coding" — and its limits
Vibe coding means building by describing and reacting — "make the header sticky", "add dark mode" — often without reading every line. It's genuinely amazing for prototypes, personal tools and learning.
For anything with real users, money or private data, add the grown-up habits: version control, tests, reviewing changes, and never pasting secrets. We'll cover them in the last lesson.
Part of 💻 Code with AI in the free AI Bootcamp.