I wrote about the quiet ratchet recently - watching AI code well is recalibrating our standards. But code is just a special case. These are general-purpose models, scoring higher on evals across unrelated domains with every release. It would be strange if our standards only shifted inside code editors.

I’m starting to see it happen elsewhere.

The meeting

Someone in a meeting was struggling to articulate what they actually wanted. They kept drifting into implementation details - libraries, databases - without ever landing on the “why we’re here” part. Couldn’t snap out of it.

So they asked an AI to do it for them. It did a fantastic job. They just handed over the output.

The phone call

A friend called a senior executive for career advice. Halfway through, he caught himself: he was doing it because it was a human. Any LLM could have given the same advice instantly.

The human voice pats you on the back by virtue of being “a person with experience.” But that’s increasingly hard to justify.

Worse: there are follow-up questions you can’t ask a person. Social mores get in the way. The moment he hung up, he was uploading financial statements to Gemini and asking the things he couldn’t ask on the call.

The pattern

Professionals get stuck. Increasingly often, they get unstuck by talking to models.

The models are gradually becoming the smartest person I know. I suspect that’s already true for a lot of people. And once you notice it, everyone else starts to have a small model smell.