It quoted me six months. I let it run overnight. It was done before I woke up. Three theories about why AI estimates are so wildly, confidently wrong — and the one I think is actually true.

AI will happily hand you a test harness that takes six hours to run. Ask one question — can this be multi-threaded? — and it becomes twenty minutes. Why this gap keeps showing up across every model, and the questions worth asking every time.

AI can generate code, tests, and interfaces at extraordinary speed — but it cannot independently define quality. Why QA engineers and UX designers become more valuable, not less, in AI-powered software development.

For years the message was clear: STEM is the only path that matters. But the skills AI rewards most — curiosity, judgment, and learning how to learn — come from the liberal arts. Why the broadest thinkers, not the deepest specialists, get the most out of AI.