Applied AI product engineering
Notes on building AI features that are useful, testable, and connected to real product workflows.
Start with the workflow
Applied AI works best when it is attached to a real workflow: reading, drafting, classifying, comparing, reviewing, or deciding. The model is not the product by itself. The useful part is the loop around it: context, constraints, feedback, and a clear next action.
Make quality visible
AI product engineering needs evaluation from the beginning. Good systems keep examples, edge cases, and failure modes close to the code so teams can improve prompts, retrieval, tools, and user experience without guessing.
Ship the narrow version
A focused AI feature that handles one job reliably is more valuable than a broad demo. I prefer small surfaces with strong defaults, clear escape hatches, and honest limits.