Cursor review: I pay for the review loop
I switched to Cursor because I wanted the editor and the model in one window. I was tired of keeping a browser tab open next to a plugin that never matched my keybindings. I still open it first every morning. That is because Cursor focuses on the review loop—inline diffs and multi-file passes—while stability and rules still feel like an afterthought.
The UI bet worked
Cursor is a VS Code fork. My file tree, terminal, and debugger stayed where I expected them. I did not have to remap my keys or hunt for breakpoints.
I just had to learn the AI modes. Chat is for questions about a function. Inline edits are for one-line fixes. Composer is for the big refactors that touch five files at once. I used to use chat for everything, which was a mistake. Now I highlight a code block, hit the shortcut, and stay in the editor. That loop beats copy-pasting paths into a browser window.
| How I work | Where it lives in Cursor | When I reach for it |
|---|---|---|
| Quick question | Chat side panel | Reading code, understanding behavior |
| Small edit | Inline accept/reject | Renames, error handling, one-file fixes |
| Multi-file pass | Composer | Refactors, tests, several paths at once |
| Odd command | Command palette | Same as VS Code |
Cursor asks you to learn three AI entry points and stay in the buffer for review. That is a fair trade for daily edits.
Composer models are for grunt work
I use the standard Composer models for most of my day. They score lower on leaderboards than the big names, but I do not care about puzzles. I care about renaming a prop across components or writing a test stub. Composer 2.5 is a step up from the last version. It handles multi-file edits without losing the thread.
Short answers come back fast. Bigger passes take longer, but I can watch which files the tool touched instead of waiting on a terminal summary. I still read every line on auth and payments before I merge. For the boring middle of the week, Composer is worth the money because I stopped paying for the priciest model on routine chores.
| Kind of work | Model I reach for | How it has gone for me |
|---|---|---|
| Refactors, renames | Composer 2.5 | A step up from Composer 2 |
| Auth, payments | Composer 2.5 + review | I read every line before merge |
| Greenfield design | Heavier model | Where I notice the gap |
Freezes cost me sessions
Stability is the biggest problem. Cursor freezes mid-session. Sometimes the whole window goes white until I force-quit. I have had quiet weeks and then two bad days in a row where I lost a Composer thread because the app died before I could copy the reply.
I save often and commit before big agent runs. That is extra work for a tool I pay for every month. It makes me hesitate before kicking off a long edit across twenty files. I keep a second editor installed just in case. If they fixed the crashes, I would recommend it to everyone. Until then, I still open it first, which tells you how much I value the review surface.
Rules are still suggestions
I want rules that keep the agent out of my generated files. Cursor has .cursor/rules, but the agent ignores them half the time. Claude Code is better here. It has clearer examples and a marketplace for skills. Cursor feels like it optimized for the diff, while Claude optimized for the agent's behavior. I want both, but right now I only get the diff.
| What I set up | Cursor | Claude Code |
|---|---|---|
| Project rules | .cursor/rules |
Clearer examples |
| Reusable skills | Improving | Easier sharing across repos |
| Checklists | More hand-holding | Easier to bake in |
| Marketplace | Smaller selection | More ready-made skills |
If your team already standardized on Claude's skills, moving to Cursor means rewriting your setup. If you work alone and only need a short rules file, Cursor is fine. I care about reuse across repos, so the gap matters to me.
Would I keep paying?
Yes. The UI is the best place I have found to supervise AI edits. Composer handles the grunt work without me upgrading every task to the priciest name.
I would tell a friend to buy Cursor if their day is mostly reading code and merging small hunks. I would tell them to keep Claude Code installed if they need to tune agent behavior across a team. Cursor is my default because review is the job, and the product is built for that job first. I am still paying while hoping the stability catches up.