Designers make mockups. Developers look at mockups. That workflow is fine when humans are reading the mockups.

AI agents can't read Figma files. They can look at screenshots, but screenshots are ambiguous. Is that padding 16px or 20px? Is that font-weight 500 or 600? Is that gap between cards 12px or 16px? A screenshot is a visual approximation. An agent interpreting a screenshot is like playing telephone with a painting.

You need to describe UI in words. Precisely enough that the agent builds exactly what you see in your head.

This is a skill. Most developers are terrible at it because they've never had to do it. You've always had a mockup, or you just eyeballed it in the browser and tweaked CSS until it looked right.

That doesn't work when an AI agent is writing the CSS. You need to learn the language of UI description.