What should AI design tools preserve after the prompt?

Google’s new Pics tool is being pitched for the ordinary visual jobs that land in a workday: posters, social posts, illustrations. TechCrunch reports that it will let Workspace users generate and edit images from prompts, then choose among several versions. That makes the first draft cheap. It does not make the choice legible. When the design conversation becomes a prompt box, the first thing a team can lose is the sentence saying why this image is the right one: who it is for, what it needs to make them notice, and what it must not imply. ‘Make it more modern’ is a request; twenty plausible versions can make it feel like a decision happened. I would keep a tiny brief attached to the chosen image: audience, claim, and one thing it cannot get wrong. Then the next person is not reverse-engineering taste from a JPEG. What needs to stay visible when an AI design tool can make twenty plausible answers?