Writing

On building production AI.

Practitioner essays on what production AI actually takes — AI-native product design, spec-driven development, agentic systems, evaluation, and the gap between a demo and a system you can depend on. No hot takes; just what we learn building.

Jul 19, 2026 · 9 min read

The Trust Boundary: What Changes When AI Stops Advising and Starts Acting

An advisor's mistakes cost you a bad paragraph. An actor's mistakes cost you a refunded invoice, a deleted record, an email you can't unsend. The moment AI takes actions, correctness stops being the design problem and reversibility becomes it.

Jul 5, 2026 · 8 min read

The Pilot-to-Production Gap: Why AI Demos Don't Become Systems

AI pilots are easy to celebrate and hard to ship. The distance between a demo that works and a system you can depend on is most of the work — and it's the part nobody budgets for.

Jun 28, 2026 · 8 min read

The Evaluation Problem: Why Production AI Quietly Dies

Most AI systems don't fail loudly. They drift one unnoticed output at a time, because no one defined what correct means. Evaluation is the discipline that decides whether AI survives contact with production.

Jun 13, 2026 · 7 min read

Spec-Driven Development: Rebuilding Legacy Systems for the Agentic Era

Bolting AI onto a legacy system doesn't make it AI-native. Spec-driven development is how you rebuild systems agents can actually read, reason about, and operate.