This started from my comment on the recent 'Language Model Teams as Distributed Systems' paper thread: [https://news.ycombinator.com/item?id=47408745].

I asked Claude to actually write/post the suggested essay, and the result drifted in exactly the ways it warns about.