My team just recently started using copilot for PR reviews.

So far I've found that 90% of what it raises is incorrect. For the stuff it actually finds, the code suggestion to fix it is almost always wrong. It will write 20 lines of code for something that's a one-line fix.

It picked up on one reentry bug on a recursive function that I don't think another dev would have spotted.

It's definitely slowing me down. I hear about AI wasting dev's time with bogus bug reports. It's now integrated into my workflow.

I've already got SonarQube and linters which finds issues the moment I introduce them. They're doing a much better job at maintaining code quality.