I'm sure most engineering is LLM-assisted already and nothing is wrong with it. It's just the one-shot vibe-coded low quality slop that spoils sentiment of this tools. Also many people are interested in what agents can build unsupervised as a test of "superintelligence".