The archive
Two posts a week on the gates, the patterns, and the cost discipline that turn AI demos into software people depend on.
Quality gates that keep AI-generated code honest
AI writes plausible-but-wrong code at high speed. These automated gates — scoring, smoke tests, visual review, a pre-push hook — stop it from shipping.
How I run a gated build pipeline for AI-generated production apps
AI can write a whole app in an afternoon. The hard part is making sure the app is correct, safe, and maintainable. Here's the gated pipeline I use to get there on the first build instead of the fifteenth.
Give your AI a memory so you stop re-explaining yourself
A small file-based memory turns a forgetful assistant into one that compounds — better output, fewer tokens, no more repeating yourself every session.
Prove the data before you write a line of app code
The most expensive AI app failures are data failures. A data-contract gate — with a reproducible anchor number — stops them cold.
The app-factory pattern for internal data apps
Stop hand-building one-off internal dashboards. Scaffold them from templates so auth, health, branding, and deploys are identical every time.
Real AI leverage without blowing your budget
How a solo operator or small team gets outsized results from AI while keeping the monthly bill boringly small.