
I started resarching trying to understand the best way to get agents to actually design/build good interfaces, and one thing stood out. When you prompt an agent to design/build a landing page, it builds the entire page and delivers it once. Designers usually design in sections.
And, when you want the agent to improve what's been done, you usually have to do it in sections to get the best results. Most "taste skills" run a full on review and gets the agent to apply fixes to the whole page or even all the pages on the project - which ends up delivering very similar results to thousands of users.
Which by default, brings everyone who has used or uses taste skills to a better version of AI slop.
English













