Blog

An Evidence-First Scorecard for AI Engineering

An evidence-first AI engineering scorecard for small teams: have the agent inspect the repo, workflow evidence, automations, and agent instructions before asking humans how mature they feel.

Native from Web — When the Spec Is Already Running

Using a working web app as the behavioral spec for an Expo React Native app: hooks vs components, when cheaper models break down, and why parity needs a written trail for the next agent.

Making SuperSubset: It's All About the Spec

Building SuperSubset taught me that the bottleneck in agentic development is the spec. The charts looked great. The tests passed. The product still failed.

Stop Thinking in Prompts. Start Thinking in Skills.

How to get more out of AI by treating skills — not prompts — as the unit of leverage. Lessons from a workshop on agentic engineering.

Hello World: I Had AI Interview Me to Write This Post

I built a system where AI interviews me instead of writing for me. Did the first interview hands-free while driving. Here's how it works.