Hey Cursor team,
I work on a large, enterprise codebase and use Cursor daily. I noticed that the composer agent generated code is much more suboptimal, in almost every request, compared to the normal composer. It also makes a lot of unnecessary refactorings that are not tied to the original query for some reason.
I definitely understand, and sometimes notice, the benefits of the agent but as long as the final code output is suboptimal/wrong, in like 95% cases, I don’t see myself using it.