Technical articles crafted for a wider audience, covering the problem, context, solution, and outcome.
This post traces strange AI output to the blanks in natural-language requests, translates requirements into a formal language of space, laws, and observations, proves them in Bend, and derives equivalence-partition and boundary-value tests with fast-check, for beginners
How to review an AI that produces different code every time by layering deterministic instruments — tests, linters, scripts — and splitting the work into small reviewable units through modularity, explained step by step
Why AI-written code goes quietly wrong, and the three pillars that fix it: reproducing correctness as tests, engineering the harness, and adding self-feedback plus an independent reviewer