I wrote in What Linear Did with Taste that taste is judgment built from proximity. Taste Cannot Be Delegated sharpens that with one line I’ve been repeating to anyone who’ll listen:
Preference describes what people choose. Judgment explains why something is the right choice.
Models are extraordinary at preference. They’ve learned from millions of examples what people tend to like. Ask for an onboarding flow and you’ll get a very preferable one: five steps, a progress bar, friendly illustrations.
What they can’t tell you is that your users wanted to skip onboarding and start working.
Organizations made this mistake long before AI did. Put a design in front of a committee and you get preference aggregation. Everyone weighs in, ownership disappears, and the strongest idea slowly becomes the safest one. A model trained on what people liked is a very fast committee.
That’s also why “do research” can’t mean “ask users what they want.” Research hands you the context judgment needs to decide what deserves to exist, not the answer itself. Skip it and you don’t lose taste in some abstract way. You lose the raw material taste works on.
Thought is a requirement, not a finishing touch
I mean thought the way people use it when they say you can tell someone put thought into this. Not thinking. Care you can feel.
Most product teams think about quality in layers. First you get the function right, then you make it reliable, and then, if the sprint has slack, you add polish: the delight layer, the touches that show someone cared.
That’s backwards.
Users don’t experience your product as a stack of concerns in the order you built them. They experience it as one thing, in one moment. If that moment feels functional but indifferent, clearly designed for the workflow rather than for them, the relationship ends. Not with a complaint. With a quiet, permanent decision to stop coming back.
You can’t put thought into a moment you’ve never seen. Thought has little to do with confetti. It’s the product anticipating the thing you were about to need, and you only know what that thing is by watching people get there.
Also in September: Taste, Thought, and Trust Don’t Come From the Model · Trust Is the One Nobody Names · The Four Cs Before the Prompt
Daniel Arevalo is a Product Architect and strategic product design leader. He helps product teams turn AI capability into trusted, adoptable workflows, with a background in industrial AI, IT/OT cybersecurity, and high-stakes enterprise software for critical infrastructure.