This is really smart. So,
Let's say we have a UI library. Agents use it and mis-use it.
How do we make that better? Evals!
Create a large set of example UI prompts.
Run the AI generation with that library.
Identify the bad bits.
Now write AI guidance to avoid those bad bits.
Run them again.
Repeat until it looks consistently good.
# Jun 29, 2026