Not everyone wants to work in a terminal. Editor-first tools put the agent inside the place you already read code.
What you get
Changes in place. The edit appears in the file, in context, with the surrounding code visible. For reviewing a subtle change, that is genuinely better than reading a in a terminal.
Familiar ground. If you already live in an editor, there is almost no new surface to learn. The agent is a panel next to code you already know how to navigate.
Fine-grained control. Accept this hunk, reject that one, tweak the third by hand. The granularity is natural in an editor and awkward in a terminal.
What you give up
The terminal as home base. Not the ability to run commands — Cursor’s agent runs your tests and installs within the approval rules you set, and Cursor also ships a CLI. What changes is where the output lands: in a panel beside the file you are reviewing. Checking out a , inspecting a log, deploying — if that is most of your day, the editor is one more window between you and it.
Composition. Terminal agents chain with everything else in the terminal, because that is what a terminal is for.
Quick check
You want the agent to make a change, run the tests, read the failure and fix it — all on its own. Which shape supports that?
Terminal or editor?
Claude Code vs is the head-to-head for exactly this choice.
Where Grok and Kimi fit
Both are model families, and both now ship their own terminal agents too: xAI has Grok Build, and Moonshot AI has Kimi Code CLI. So there are two ways in — the model as the engine inside a client you already like, or the vendor’s own agent in the same seat as Claude Code. Either way they show up in this conversation for a specific reason: cost and availability. If a frontier tool is out of budget or unavailable where you are, a cheaper or more open alternative is not a compromise — it is the difference between learning this skill and not.
The workflow does not change. You still give context, review a diff, iterate. What changes is how much correction you do per task, and that is worth measuring for yourself rather than taking anyone’s word for.
The point of this whole course
Every tool here is the same shape. Learn the shape — give context, ask for a plan, read the diff, iterate — and the specific tool becomes a detail you can change in an afternoon.
What to do now
Pick one and use it on real work for two weeks. If you have no strong preference, the Claude Code course on this site takes you from install to a shipped project, and everything you learn there transfers. Already settled on another terminal agent? Codex from Zero and Gemini CLI from Zero cover the same first steps for those two.
Keeping up without re-reading comparisons
This course deliberately avoids anything that expires — model names, prices, context sizes, benchmark scores. That leaves a gap: sometimes you genuinely want to know what a vendor shipped last week.
We keep that in one place, [agents-compared](://github.com/xpressmike/agents-compared) — an open repository with a page per tool, written to the same rule as this course, plus a machine-maintained list of each vendor’s latest announcement. A weekly job checks every link and flags any page that an announcement seems to contradict, so the list either stays true or admits that it does not.