When doing AI tasks that involve project wide changes eg change a file path in code and there maybe hundreds of occurrences, Cursor uses a very slow search and edit file approach rather than leveraging the existing native find and replace functions that are fast and do not demand AI for each replacement.
Consider exposing the native implementation as a callable tool to the models.
Currently this performance makes large scale code tidying and refactoring so painful that it easier to no use AI when this type of simple task would appear to be a perfect AI candidate,