Most AI product pitches talk about speed. "Save 30% of your time." "Fewer clicks." "Instant answers." But what almost nobody measures is the thing that actually decides whether a tool is useful in real work: What does this do to the person's cognitive load?

Most "AI assistants" look the same on the surface. A chat box. Some canned prompts. A promise that it can "help with anything." That's tolerable for drafting emails. It's reckless in law, finance, and medicine.

Most teams still treat generative models like vending machines. You type a clever prompt, hit generate, get something that looks surprisingly good, drop it into a deck, and call it a win. That works for experiments. It does not work for production.