Cursor Tab

The Claim

Cursor Tab is supposed to be the smartest autocomplete in the game. Multi-line predictions, context-aware suggestions that understand your entire codebase, not just the current file. The pitch is that it feels like pair programming with someone who actually read the docs.

What I Tried

I gave it two weeks on a Next.js project I was already building. Forced myself to use the Cursor editor full-time, which meant leaving neovim. That alone should tell you how much I wanted this to work.

The tab completions are genuinely good. It predicted multi-line function bodies accurately maybe 60% of the time, which is better than Copilot's hit rate. The "next edit prediction" feature, where it guesses where you'll edit next and pre-fills the change, is a neat trick that occasionally saved real time.

I tested it on TypeScript interfaces, React components, and some Solidity test files. TypeScript was its strongest suit. Solidity was a coin flip.

What Surprised Me

The latency. Cursor Tab is noticeably faster than Copilot for inline suggestions. There's almost no flicker or delay. Whatever they're doing on the inference side, it's snappy.

What also surprised me is how much I missed vim motions. Yes, Cursor has a vim mode. No, it's not the same. The keybind conflicts alone cost me probably 30 minutes over the first few days. `ctrl-a` doing something unexpected, marks behaving differently, macros that just didn't work right. Death by a thousand cuts.

Who It's For

If you're already in VS Code and using Copilot, Cursor Tab is a legitimate upgrade. The multi-line completions are a real step forward, and the editor is polished enough that the switch from VS Code is painless.

If you're a terminal-first developer with a tuned neovim config, the math doesn't work. You're trading a 10-15% improvement in autocomplete quality for a 30% hit to editing speed while you relearn muscle memory in a new environment.

Verdict

Dropped. The completions are impressive in isolation, but I don't evaluate tools in isolation. I evaluate them inside my actual workflow. And inside my workflow, switching to a GUI editor is a net negative. The marginal autocomplete improvement doesn't offset the friction of leaving the terminal. I'll revisit if they ever ship a headless mode or a neovim plugin.