There are two main places an can live, and picking the wrong one for your habits is the most common reason people bounce off these tools.
The terminal agent
You open a terminal in your project folder, start the agent, and talk to it there. It reads files, proposes changes, applies them, and can run commands — your tests, your , git.
That last part matters more than it sounds. An agent that can run your test suite, read the failure, and try again is doing something qualitatively different from one that can only suggest text at you.
The cost: you are in a text interface. No file tree, no click-to-open. If the terminal is unfamiliar, that is friction on top of learning the agent. This is exactly why the Terminal Basics course exists. (Some terminal agents also ship an editor extension or a desktop app — Claude Code has both — but the terminal is where they are most at home.)
The editor agent
The agent lives inside your code editor. You see the file, it edits the file, changes appear in place with accept and reject buttons.
The advantage is context — literal, visual context. You see what changed, where, surrounded by the code around it. For people who already work in an editor, it is a much smaller step.
The cost is not reach any more. Editor agents run commands too — Cursor’s agent will run your tests or install a package, within the approval rules you set. The cost is where things land. You review file by file, hunk by hunk, and the terminal — where the test output, the and the deploy log print — is a panel beside your work rather than the place your work happens.
Quick check
Your agent changed five files and you want to know if anything broke. What actually tells you?
The third shape: agents that work somewhere else
Some setups run the agent on a remote machine. You give it a task, it works on its own, and you get back a branch or a pull request to review.
This is a genuinely different mode. There is no back-and-forth — you write the task well or you get the wrong thing. It suits well-specified, self-contained work, and it suits people who already review other people’s code for a living. It is a poor first tool for a beginner, because the feedback loop that teaches you is exactly the part it removes.
If you are undecided
Start with a terminal agent. Not because it is better, but because it teaches you more: you see the commands, the failures and the fixes, and that knowledge transfers to every other tool. Moving to an editor agent afterwards is easy. The reverse is harder.
What this course does next
The next level goes tool by tool: Claude Code, Codex, Gemini , Cursor, and where Grok and Kimi fit. Each one gets the same treatment — what it is shaped like, what it is genuinely good at, and what will annoy you.