The Concept Is the Constraint

When you ask an AI to design something, it reaches for the average. Not because it is lazy — because it is correct. The average is what "good enough" looks like across the distribution it was trained on. Chat interfaces look like chat interfaces. Onboarding flows look like onboarding flows. The AI is not wrong. It is producing the most statistically defensible answer.

The problem is not the AI. The problem is the absence of something the AI cannot invent: a reason to deviate.

A strong product concept is not decoration. It is a constraint — a force that pulls the design away from the mean toward something specific. "This should not feel like a chat interface" is not a stylistic preference. It is a structural instruction that changes every downstream decision. The AI can execute that instruction. It cannot generate the instruction itself.

This is why AI-generated software tends toward homogeneity at scale. It is not a failure of capability. It is a failure of input. When the concept is absent, the AI defaults to the center of the distribution. When the concept is present and specific, the AI can be pulled away from that center — and held there.

The question is not whether to use AI in design. The question is whether you have something worth constraining it with.