antirez (@antirez) on X

Fascinating that this is where we're at. I still look at the code and there is still a general perception that this matters, but it's not really evidenced based. It would be fascinating to see some experiments run to compare results and make the comparison less about vibes and preconceived ideas. There is no doubt in my mind that if you give an agent the right spec, the right tools and loops that it can implement that it can build some top tier stuff. Frontend design and taste I'm still dubious about, I've used Opus 4.8 with a browser QA skill multiple times and had it pass work that when I look at I immediately notice things that don't make sense. Nonetheless if the correct guardrails (hard ones where possible) are put in place, a huge amount of work can be largely automated with clear and good direction.