Long-form writing on building AI products — what's working in production, what we've stopped doing, and what we're paying attention to. New entries land every few weeks. No newsletter, no pop-ups.
Why the interesting work in a legal-review copilot was everything that happened before the model saw the prompt.
What we learned rebuilding a brokerage's weekly ops into a single AI-assisted console — and what we almost got wrong twice.
A practical argument for writing your evaluation harness in the first week, not the last. Includes the scorecard we actually use.
The journal runs at studio pace, not content-mill pace. We publish when we've learned something worth writing down — usually after shipping it.
A new entry every few weeks. No marketing, no scheduled cadence — only when there's something worth writing down.