Designing reliable AI agents starts at the tool boundary
Identity, authorization, durable state, and recovery determine what an agent can safely do after it chooses an action.
ESSAYS / ENGINEERING & ARCHITECTURE
Thinking in systems.
Technical writing about the decisions that make AI systems useful, reliable, and possible to operate. Agents, retrieval, evaluation, security, and the engineering between them.
Subscribe via RSS ↗Evaluation is infrastructure, not a release-day checklist
A useful evaluation system connects versioned evidence, explicit failure categories, and release decisions across the life of an AI application.
WHY I WRITE
Make the reasoning explicit.
Writing is a way to examine a decision: the problem it solves, the assumptions it depends on, and the cost of choosing it. These essays connect that reasoning to public engineering work.