DEV Community

Carlos Casalicchio
Carlos Casalicchio

Posted on

We published the article most "AI made our code better" posts skip — the honest

We published the article most "AI made our code better" posts skip — the honest comparison.

Results:

  • Correctness: tie (~90% vs ~95%)
  • Quality (blind test): skills won both tasks at high confidence
  • Cost: 24% more tokens, 33% more time

The insight: "Correctness is solved. Craft is not." The durable gains come from verification loops, not longer instruction files.

Full article: https://splatdev.com/blog/ai-agent-skills-for-front-end-the-gains-the-gaps-and-an-honest-comparison/

Top comments (0)