Every time I use Agent and then open the PR, Bugbot finds several problems that were generated by Cursor itself or that should at least be identified by Agent, since I am always ‘talking’ to it.
Why does this happen?
What model are you using for Agents?
I never change the model manually, I always use Auto.
yeah that would be the problem, top of the line models find problems better and also context changes, code review reviews the diff of changes while agent works with entire codebase and even if you would use the same model:
- agent - generates
- review - finds some problems - fixes
- review (again) - find more problems
llms are not deterministic, if you would run code review multiple times in a row without fixing anything, the actual review might and likely will differ