@dankrad Also, coming from a verification background, the tooling doesn't seem to be quite there yet. There are no vibe-proven systems in production yet, and the gap between toy projects (or math) and large production codebases is enormous.
@dankrad Have you actually had success doing this? Speaking from an engineering team that is all-in on LLM usage, we are seeing a slight productivity boost . It only takes one bug in unread slop to completely ruin you, so manual review is mandatory unless you are willing to get rekt.