The part of Cursor I still pay for isn't the editor

Alex Yumashev cancelled Cursor, and I read the whole post nodding. Then I noticed I was nodding at the wrong parts.

His complaints are all real and all about the editor. Cursor rides a VSCode build that's seven or eight months old, so extension updates refuse to install. Every other update boots you back into the Agent window, and the setting to stop that keeps getting moved and renamed. Bugs he reported a year ago are still open. Support replies read like a bot dodging the question. He's leaving because Cursor stopped being a good IDE, and he says he'll come back "the day they decide they're an IDE company again."

I'd get there too, but from the opposite door. For the work that actually matters to me, I stopped using Cursor as an editor a while ago. So the question his post kept circling never quite got answered: if the value is Claude or Codex, and you can run those anywhere, what is Cursor's remaining marginal utility? Is the editor the product, or a shell around someone else's model?

The line under all the complaints

Yumashev buries the real argument in the middle of his own post. When he actually needs an agent, he writes, he'll reach for Codex or Claude Code, and the joke is he can do that from inside Cursor. Once the model makers will sell you their top tier directly, a second subscription that just marks up the same models is hard to justify.

That's the whole thing. The VSCode lag and the Agent window and the four-month CSS bug are symptoms. The disease is that once the model is the value, the editor around it is a wrapper you're paying a second subscription for. His complaint is that the wrapper is cracking. Mine is that I already stopped reaching for it.

What already left the editor

The agent work isn't in Cursor for me anymore. It's Claude Code, which I run in its own app more than the terminal, and I let it drive gh directly: opening PRs, reading review threads, replying in-thread, splitting a sprawling branch into stacked PRs. That loop has nothing to do with an editor. It runs against the repo and GitHub, and it runs the same whether Cursor is open or not.

So the pivot Yumashev is angry about, more agent and less editor, isn't the thing that pushed me out. The agent I want just doesn't live in the editor at all. When Cursor puts the Agent window in your face on every launch, it's trying to be the surface for exactly the work I already do somewhere else, with less lock-in and closer to the model.

The part I still pay for isn't an IDE either

Here's the part that makes me feel like I'm arguing against my own subscription. The one piece of Cursor I still get real value from is Bugbot, its automated PR reviewer. And Bugbot doesn't run in the editor. It runs on pull requests, on GitHub, billed on its own line. It's caught things a local tsc pass never could: an upload that would have thrown against the real Blob API on the very first run, a prune step that would have deleted the backup it just made.

Notice what that means. The one thing keeping "Cursor" in my monthly spend is a product that isn't the IDE, doesn't touch my editor, and would work exactly the same if I wrote every line in plain VSCode. I'm not paying Cursor to be my editor. I'm paying a company called Cursor for a reviewer that lives on GitHub. The drift Yumashev objects to, away from a code-first editor and toward an enterprise agent platform, is for me the reason the subscription survives at all.

So where does the editor still earn its keep?

I don't want to overshoot this. The editor isn't zero. There's a class of edit (rename this across the file, restructure this block, the change I can already see in my head) where tab-completion and an inline Cmd-K edit are just faster than describing the change to Claude Code and waiting for a diff. The friction of switching to prose isn't worth it for a ten-second change. That's genuine, and it's the reflex I haven't dropped.

But that's a narrow band. It's the small, local, I-know-exactly-what-I-want edits. Everything with real blast radius (multi-file features, migrations, anything where I want the model to gather context and reason across the repo) left the editor for Claude Code, and the review of all of it left for the PR. What's in the middle, the tab-complete and quick inline edit, is exactly the stuff plain VSCode plus an extension does about as well now. Which is Yumashev's other point: the features Cursor introduced are everywhere now, and they're everywhere because Cursor got there first.

Worth it depends on what you're buying

"Worth it" was never a yes or no. It depends on what line item you think you're paying for.

If you're buying an editor, Yumashev is right to cancel. The editor is stagnating, and the good parts are commodity now. If you're buying an agent, you're better off buying it from the people who make the model, closer to the source, with less between you and it. And if what you're actually buying is a specific non-editor product that happens to have Cursor's name on it, Bugbot in my case, then the IDE debate is beside the point. You're paying for the reviewer, not the editor, and you should be honest with yourself about that on the invoice.

The switching cost is the tell. Ask what you'd actually lose by closing Cursor tomorrow. For me it's a few seconds on small edits I could relearn in plain VSCode by Friday, and nothing else, because the agent already lives in its own app and the reviewer is already on GitHub. When the switching cost of an editor is measured in muscle memory rather than capability, the editor was never really the product.

And even that last line item isn't safe. Bugbot is a reviewer I happen to like, not one I'm locked into, so it competes on the only thing left once portability is a given: the best review quality per dollar. The next thing I'm doing is putting CodeRabbit and Greptile on the same pull requests with the same answer key, the bugs I already know are sitting in that history. If one of them catches as much for less, the last reason Cursor is in my budget leaves with it.

Yumashev cancelled because Cursor stopped being an IDE company. I still pay because it stopped being one for me too, and started selling the one part I actually use. I plan to cancel that part as well, the day another reviewer wins the same answer key for less.