I think you've captured how I feel about it too. If I try to go beyond the scopes you've described, with Cursor in my case and a variety of models, I often end up wasting time unless it's a purely exploratory request.

"This package has been removed, grep for string X and update every reference in the entire codebase" is a great conservative task; easy to review the results, and I basically know what it should be doing and definitely don't want to do it.

"Here's an ambiguous error, what could be the cause?" sometimes comes up with nonsense, but sometimes actually works.