I really do like Cursor’s tab completion as it is really fast and perfect for refactorings. When it comes to coding new stuff, i disable it, as it will hallucinate at high speed while having no clue, what i actually want to do. The tab completion has always been very fast and overly aggressive in suggestions. But that was fine, as long as i disable it, when there is no context to work with.
Now it seems to me, that the tab completion has significantly dropped in it’s predictive quality even when there is context. I am refactoring code and get a lot of wrong and really bad suggestions working in Dart. This aligns to my experience with Composer 2, which is overthinking and doing way too much.
Hey, thanks for the feedback on Tab. To help us look into this properly, we’ll need a bit more detail:
Your Cursor version Help > About or Cursor > About Cursor
A couple of specific examples of bad Dart suggestions, a screenshot or a code snippet showing expected vs actual output
Roughly when you noticed the drop, like the last few days, after an update, etc.
About Composer 2 “overthinking”, that’s separate from the Tab model, they’re independent. But if you noticed both getting worse at the same time, mention it and I’ll pass that along too.
I can’t share a timeline for Tab model changes right now, but if you can send concrete examples, I can forward them to the team that owns Tab.