It writes better code than me. The most serious point is that it's still a proof of concept; I wouldn't trust money to such a thing, but nowadays that's nuanced: it's not "i wouldn't trust it as much as myself", it's "I wouldn't trust it without some review by humans and a lot of review by other LLMs".

For now I'm writing mostly the papers of these PoCs and then having the bots do all the hard graft :)