I am in a workspace which has 10s of thousands of files and directories. That has never been a problem.
I noticed a HUGE slowdown in responses (2 seconds → 30-60 seconds) since 0.49.x. I eventually isolated this to including the project structure, which presumably crawls the whole workspace, either every query, or the representation is so non-compact that the LLM response takes a LONG time.
Thing is: I really like the idea of including project directory structure! Here is an example request ID so the cursor team can try to improve this: 3cd8b328-5578-4386-84a2-db3f3fdd797b (this took around 30 seconds to run a one-sentence question).