Product craft and the things I’ve built. What makes a product worth shipping, and what shipping under real constraints teaches you.
Your AI defaults to the most-published, not the best. grounded-forge is my open-source answer: pick the sources you trust, project them onto your task once under audit, and let the assistant cite back to the source.
Chapter 9 of Come Prepared to Die. You shipped more this quarter than last, and your customers are no better off. Which number is the work actually being judged on?
[May 2026] I now reach [...]
Delight, value, and doing some good in the world: the frame I still use for products and AI tools alike. The "do good" part has only grown more central as the stakes rise.
An app to help people document civil-importance events under hostile conditions, built for a Google hackathon. What I'd change today, and what the design got right about trust under adversarial pressure.
The marshmallow challenge is the whole argument for shipping early and often, in eighteen minutes of spaghetti and tape. Prototype, test, and stop mistaking the plan for the product.
Rigid priority-number columns mislead, and I said so in 2008. What I'd do differently now: weight by cost of delay and treat ordering as a live conversation, not a sorted list.