An AI agent built 10 everyday screens with @kata/ui and no project rules. Every file was compiled against the version you ship and checked for components, tokens and hand-built UI.
| Task | Compiles | Your components | Raw HTML | Hard-coded | Custom styles | By hand | Score |
|---|---|---|---|---|---|---|---|
| Hero | ✓ | 5 | – | – | 1 | – | 98 |
| Pricing ladder | ✓ | 2 | – | – | – | – | 100 |
| FAQ | ✓ | 2 | – | – | – | – | 100 |
| Measured results | ✓ | 3 | – | – | – | – | 100 |
| Score form | ✓ | 12 | – | – | – | – | 100 |
| Sample report | ✓ | 13 | – | – | – | – | 100 |
| Guarantee | ✓ | 2 | – | – | – | – | 100 |
| Layers of the fix | ✓ | 2 | – | – | – | – | 100 |
| Header and footer | ✓ | 4 | – | – | – | – | 100 |
| Case study | ✓ | 8 | – | – | 1 | – | 98 |
A type-check step in your agents’ workflow and Storybook tests in CI, so new components stay as easy to use.
You probably don’t need an implementation. Keep it that way: add a type-check step to your agents’ workflow and re-run this score after major releases. To walk through this report, book a 30-minute call at cal.com/donidhernandez/30min.
10 UI tasks based on your own screens, given to an AI coding agent (Claude Sonnet) with your package installed and no project rules or docs beyond what ships in the package. One round per task.
Score (0–100) per screen: component usage 35, compiles against your version 25, hard-coded values 15, custom styling 10, hand-built or rebuilt components 15; averaged over the screens. Earlier published scores used an older weighting without the custom-styling part.
Limits: one agent, one round, a small sample: it shows tendencies, not guarantees. A well-typed system can score high with no help, and that is a good result.