I think this still is some kind of 'fight' between assisted and more towards 'vibe'. Vibe for me means not reading the generated code, just trying it and the other extreme is writing all without AI. I don't think people here are talking about assisted : they are taking about vibe or almost vibe coding. And its fairly terrible if the llm does not have tons of info. It can loop, hang, remove tons of features, break random things etc all while being cheerful and saying 'this is production code now, ready to deploy'. And people believe it. When you use it to assist, it is great imho.