I help teams ship AI that holds up in production.

Most of my work now is with teams putting AI agents and retrieval systems into production, and finding that the demo that impressed everyone doesn't survive contact with real traffic, real data, and real edge cases. I help them close that gap: the evaluation, the retrieval quality, the failure handling, and the architecture decisions that separate a system from a demo.

I came to AI from fifteen years of building production software: distributed systems, Java, Kubernetes, the unglamorous machinery that has to stay up. That background is the whole point: I've spent a career on the things that break at 3am, so when I say most agents in production are demos with extra steps, it's from having debugged the difference. Today I'm an engineer at a global logistics company and help lead its AI developer community.

I wrote The AI Agent Playbook and a series on agents in production for the same reason I write everything here: to pin down what actually works, past the tutorial. The writing is read by practitioners building the same systems, and it's where the arguments behind my work get made in full.

Right now I'm building a production RAG knowledge assistant, the kind of project this whole site is about. When it ships, it becomes the case study.