↑
logo

All code is slop (except mine) [video]

Posted by jebarker |2 hours ago |1 comments

sunaurus an hour ago

When agents keep creating problems for themselves to solve at a massive scale, "PRs merged" is an even more terrible metric to track productivity than it is normally.

I've also been letting an army of agents merge hundreds, sometimes thousands, of PRs a week for a while now on several side projects. I keep reviewing their output intermittently, and I keep finding absolutely horrifying things.

It's just an absolute mess. I believe in my case, the useful output of the agents is decreasing week by week. The vast majority of work they are doing is just a complete waste of time as they go around in circles, feed on bad output generated by previous agents, and spread this bad output throughout the codebase (as the presenter in the video actually pointed out as well).

I've actually been implementing the same types of constraints mentioned in the presentation, and definitely even gone further in some ways. Overall, I fully agree with the levels on her pyramid, and my own priorities for constraints are really similar. Still, despite that, the level of tech debt that accumulates when working like this is unbelievable, and I seriously doubt it's a good idea to work like this for long-term product health.