OpenAI’s new GDPval benchmark shows AI can produce spreadsheets, decks, and care plans that domain experts sometimes prefer to human work. But MIT’s State of AI in Business report found that 95% of enterprise AI pilots still fail to scale. This seems like a paradox: the capability is there, but adoption lags badly. I explored this in a recent essay (linked), but I’m curious what HN thinks: where’s the real constraint?