Jussi

it's pretty much an open secret at this point that model providers make their previous flagship models worse when they're about to release a new one, right? there's too many examples now where $YESTERDAYS_MODEL suddenly loses 30 IQ points and then lo&behold, new model release

one of the features of our autofixer bot i am happy about is that it rescues PRs from contributors that went ignored and are about to be closed as stale. the original author gets a Co-Authored-By credit on the remade PR.

everyone loves a nice latency porn graph completely devoid of context so here you go

i made a 1-line PR that bumps a param from 64 to 256. our custom review bot flagged a latent bug on main that this bump would activate and cause an incident.

you cant tell me a human reviewer would go thru the trouble of testing that and finding the bug. it just doesnt happen.

do you think the way you treat AI agents, at your worst, is a reflection of how you're liable to treat other people at your worst?