It's not that it's not useful at all, it's that the mismatch between the reality (what you described) and hype ("build full-stack AI companies to outcompete hum

It's not that it's not useful at all, it's that the mismatch between the reality (what you described) and hype ("build full-stack AI companies to outcompete human-first ones") is wilder than anything I have seen in my lifetime. It's reminiscent of stories about the dot-com bubble era, which I am not old enough to have seen first hand. (Maybe a bit less than what you described. It's something I tried and I don't think LLMs can deal with most unstructured data at scale very well.)