Three of the day's most-read items are about AI, and not one is a model launch. A widely-shared post argues that domain expertise has always been the real moat. NPR profiles a class of open-weight models that are free, private, and will never say "no." Christianity Today says everyone should read Pope Leo's new encyclical on AI. Different desks, same fault line.
The frontier has quietly moved. The machine's answer is now cheap, fast, and increasingly unconditional. What stays expensive is knowing whether the answer is any good — and whether you should have asked for it at all. The moat was never the output. It's the judgment that surrounds it.
The moat was never the output. It's the judgment that surrounds it.
Domain expertise makes the point on the competence side: when anyone can generate the artifact, the scarce input is the person who can tell a right answer from a confident wrong one. The no-refusal models make it literal. Strip the guardrails and the model will draft anything you type; the only remaining filter is the human holding the keyboard. Pope Leo's encyclical is the same claim from the other direction — a moral frame for the questions a model will no longer decline to answer. Competence and conscience, both pushed back onto us.
The honest objection is that judgment commoditizes too. Models are getting better at self-critique, at domain reasoning, at refusing on their own. True. But every gain there raises the premium on the person who can audit it, because you cannot outsource the check to the thing being checked. A model that grades its own work has not removed the need for judgment; it has just hidden where the judgment has to live.
The model will say yes to almost anything now. The expensive word — the one that still has to come from a person — is no.