Agents that know when to stop
An agent that cannot stop is a liability. How we give every agent a step budget, a typed stop reason, a way to hand off to a human, and evals that check all three.
AI
Agents, evals and the craft of putting models into production.
One waymark when we find something worth marking.
Engineering write-ups, AI field notes and product decisions from Kairn Labs. Roughly twice a month. No tracking pixels.
