Cursor or VS Code with Copilot: is leaving the mothership worth it?
I have bounced between Cursor and plain VS Code with Copilot three times in two years, which should tell you the answer is not obvious. Each move felt right for about four months, and then something small nudged me back the other way. Where I have finally landed is a split, not a winner, and I think the split is the honest answer for most working engineers.
what the fork actually buys
Tab and Composer are wired into the editor rather than bolted on as an extension, and you feel it in the small stuff: the prediction fires faster, the agent has a tighter grip on the editor state, the whole thing behaves like one product rather than a host plus a plugin. Cursor's pitch is integration. Stock VS Code with Copilot has closed a lot of that gap, and for chat-style help it is genuinely close now. For the inline prediction specifically, Cursor still feels a clear step ahead to me, and I get into why in what still makes Cursor the one to beat.
what staying on the stock editor buys
The other column is where it gets interesting, because it is not really about features at all. Staying on plain VS Code means no fork-update tax, so I am never waiting on Anysphere to catch up to an upstream release my extensions already assume. It means whatever my client's IT department has already blessed, since a bank is not going to let me install a forked editor on a managed laptop no matter how good the autocomplete is. And it means the extension ecosystem behaves exactly the way upstream ships it, with none of the fork-shaped surprises I have burned an afternoon on before.
The real question is whether you can afford to run a second editor that a single company controls.
where I landed, split by whose repo it is
So my rule now is boring, and it depends entirely on whose machine I am sitting at. For locked-down client work I stay on stock VS Code with Copilot, because the org already trusts it and I am not going to spend political capital fighting to install a fork on a fintech's hardware. For my own projects, and for greenfield client repos where I set the rules, I run Cursor every single time, for the Tab. I keep both installed on the M2 Ultra and I have stopped feeling bad about it. If you want the fuller matrix of who should pick what, the tools compared is where I put the whole thing.
the lock-in I actually think about
The quiet risk underneath all of this is control. Cursor is one company's fork, and if Anysphere changes direction or pricing in a way I do not like, I am standing on their editor, not a neutral one I can easily walk away from. Copilot rides on top of VS Code, which Microsoft is not about to fork out from under me. That asymmetry is not a dealbreaker, and I still run Cursor daily on my own work, but it is the same instinct that makes me keep a vendor-neutral CLI in the toolbox as insurance, which I argued for in open CLIs versus vendor lock-in. Bet on the fork with your eyes open, and keep an exit.