I mean if I were writing something like cursor my feedback system would be did the user move on to the next problem without making changes, that looks like success, if they changed the code in the area manually X amount before asking the next problem that would be failure.

I wouldn't be waiting for people to click thumbs up/down to start checking requests as success or not.

If you want to harm it's training you need to use it but teach it to solve the wrong things, or pretend to use it but actually do your work manually. If you're using it then it's already over if your mindset is this is going to replace you.