Lesson 2 of 6
How Claude Code works
7 min read
Cursor lives in an editor. Claude Code lives in your terminal. Strip away the interface — what's actually running underneath?
An agent in your terminal
Claude Code is a coding agent that runs where your code already lives — the command line. There's no special editor UI; it reads your files, runs your shell commands, and edits in place. Underneath, it's the same shape as any agent: give it a goal, and it runs think → act → observe in a loop until the job is done.
Claude Code is the read-plan-edit-run loop with a terminal as its home. Its 'act' is a shell command; its 'observe' is that command's output.
Same loop, different tools
The scene shows a generic agent booking a table, but the mechanism is identical to what Claude Code does with your repo. Swap the tools and you have it: instead of a booking API, its actions are read this file, run this command, apply this edit. It's an agent, not a fixed workflow — nobody scripted the exact steps; it decides them turn by turn.
The loop is universal; only the tools change. Claude Code's tools are your file system, your shell, and your test runner.
Because its 'act' is a real shell command, Claude Code can do anything your terminal can — run tests, grep, git, install packages. That power is also why it asks before running commands that change things.
What makes it Claude Code
- —Terminal-native — it works in the shell, not a bespoke editor.
- —Real commands — 'act' means running an actual shell command and reading its output.
- —Agentic — it decides the steps toward your goal; you don't script them.
Underneath its terminal interface, what is Claude Code actually doing?
Continue in the app
Take the whole How AI Coding Agents Work course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.