We just published a detailed investigation into whether AI agent skills can push front-end from "competent" to "genuinely beautiful" — with blind testing to prove it.
Key findings:
- Built 8 skills (7 knowledge + 1 process)
- First evaluation: tie on correctness (we were measuring the wrong thing)
- Blind test: skilled agent won both tasks at high confidence
- The process skill (render-critique-refine loop) was the breakthrough
The article covers the full methodology, results, and what we changed based on the findings.
Full write-up: https://splatdev.com/blog/how-ai-agent-skills-produce-measurably-better-front-end-blind-tested/
Top comments (0)