July 2, 2026
Why green evals still ship broken agents
A test suite proves your agent handles the cases you imagined. Here's what that misses — and how reading production closes the gap.
Read more →The Hone blog
Field notes on making production AI agents better every day.
June 18, 2026
Intent is the unit of agent quality
Accuracy metrics hide the real question: did the user get what they came for? A look at scoring agents by intent instead.
Read more →June 3, 2026
Anatomy of a self-improving pull request
How Hone turns a recurring failure into a diff, replays it against history, and hands it to a human to merge.
Read more →