Skip to main content
/kl
  • Writing
  • Speaking
  • Résumé
← All topics

Topic

foundation-models

1 article on this topic.

  • July 19, 2026 8 min read

    My Agent Kept 'Improving' My Prompts. The Eval Harness Said Otherwise.

    Agentic development without evals is approving diffs on vibes. How a small Swift eval harness caught real prompt regressions, killed a temperature theory, and told me when to stop iterating.

© 2026 Kyle Lee

  • RSS
  • GitHub
  • YouTube
  • Twitter