We're currently wasting a lot of time cleaning up slop created by overusing AI. The loop is: developer uses AI to create a huge amount of code very quickly. But the developer didn't understand the underlying system very well so the specs generated by the AI contain a lot of faulty assumptions and the code was then built on those assumptions. So you have features that probably shouldn't have existed in the first place, but because they do exist people have provided feedback on them, and the developer then uses the AI to implement that feedback, but still doesn't understand the system so much of the code for the revisions is still wrong. So that individual developer saw a huge productivity increase, in the sense that he generated a vast amount of code really quickly, but the team as a whole saw a productivity decrease.