Cursor's Tab: Why One Keystroke Beat the Chat Box
Cursor's multi-line tab prediction won developers not by being smarter, but by costing one keystroke instead of a sentence.
What I noticed
The feature developers talk about in Cursor is not the chat sidebar. It is Tab. You write a line, and Cursor predicts the next edit, often several lines across multiple locations, and you accept the whole thing by pressing Tab once. It feels like the editor is finishing your thought before you finish typing it.
Why it works
Every AI feature has an invocation cost: the effort to ask. Chat is expensive. You stop, switch context, phrase a request in full sentences, wait, then read and paste. Tab completion costs nothing. You keep typing, and the suggestion appears in the flow you are already in. Accepting it is one key you were already resting a finger near.
Cursor understood that in coding, intent is already on the screen. The surrounding code, the file you just edited, the function signature: that is the prompt. You do not need to describe what you want because you are in the middle of doing it. By reading that context and predicting the edit, Cursor removed the entire act of asking. The model does not have to be dramatically better than a chat assistant. It has to be good enough while costing one keystroke instead of a paragraph.
The tradeoff
Cheap invocation makes wrong suggestions cheap to accept. A confident, plausible, subtly wrong completion slips in with the same single Tab as a correct one. The interaction that makes Cursor fast also makes it easy to absorb bugs you would have caught if you had been forced to type the line yourself.
What I’d steal
Lower the cost of invoking your AI before you raise its intelligence. If using the feature requires a context switch and a written request, most users will not bother, no matter how good the answer is. Find the action the user is already taking and attach the prediction to that.