I’ve noticed with recent projects that the application / merging of the code returned from the LLM (usually claude 3.5 sonnet) is more often than not quite poor. I can see that the LLM will add comments, e.g. “add near the top after blah” and cursor via whatever it uses to work out where to put it, ends up putting it at the bottom of the code. Or things like modifying existing functions, the results from the LLM look good, telling cursor to merge in the appropriate place, then cursor just sticks it all at the end as new code.
I’m using 0.43.6. It just seems rather bad recently when applying and it’s rather frustrating!