When AI came in full force early this year, I got a lot of junk code contributions. Lots of silly changes. Really annoying. As of two months ago, it is a lot better. I am still somewhat swamped with outstanding pull requests that I need to assess. I know a lot of people are frustrated as their PRs sit unreviewed. The bottleneck is that I need to understand what is going on. Why? If the AI says it is fine, why not just merge? Sometimes that is the right move. But I worry about nonlinear effects. A change can do all the marvellous things the commit message claims and still hurt the code base in ways that are not immediately obvious. Performance optimizations are especially nasty. It is quite easy to improve benchmark A while making benchmark B worse, and you do not always have benchmark B ready at hand. I can often use judgment, based on many years of experience, to assess probabilistically whether an optimization is worth having. But when the changes get complicated, I become much less sure of myself. Plus, I am often wrong. I need time to think. If it were all 'AI slop', it would be easy. But the code contributions that I have on my pile right now are likely good work. So, annoyingly, I prefer to slow things down deliberately. Thus far, it still seems sustainable, but I am preparing myself for a world where I will just deliberately and brutally dismiss a large fraction of code contributions, sometimes arbitrarily.