The study where developers lost productivity due to using AI tools should always be mentioned with caveats.

This comment by simonw for reference: https://news.ycombinator.com/item?id=44522772

1/ They tested on 16 experienced developers. 2/ They only used cursor. No mention of other tools at the time. Cursor is notoriously bad on bigger codebases, while even Windsurf (competition at the time) would have shown different results. 3/ They only allowed training for 30 mins on cursor. 4/ As simonw mentions, the ones who improved had some previous experience with cursor. 5/ Yet, the study is definitively mentioned as AI actively ruining productivity and none of the other factors.