AI systems make design decisions constantly. When a model decides whether to say '30% chance of rain' or 'unlikely to rain,' that is a design choice, and right now most teams are not controlling it. Nielsen Norman Group's latest piece argues that the core skill designers must develop is critique: the ability to encode user needs and design judgment into evaluation criteria that actually shape model behavior.
The article's most useful move is reframing the problem through the shift from deterministic to probabilistic systems. Traditional specs work because engineers implement exact behaviors and QA validates them. Generative AI breaks that contract. You cannot write a Figma spec for every possible model output. The piece forces designers to confront what replaces that contract, and the answer is not prompts or vibes.
The full article is worth reading for how it operationalizes critique as a transferable method, not a one-time audit. If your team is shipping AI-powered products and still relying on gut checks to evaluate output quality, this is the framework you are missing.
[READ ORIGINAL →]