Notes on building AI that ships.
Practical writing on agents, evaluation, automation, and turning research into products — from someone who does it. Seven notes so far, each with its figures and its sources. Subscribe to get the next ones when they land.
The moat moved.
AI coding agents made software cheap to build. What stays defensible is the invention, the patent that protects it, and the reputation you earn by publishing what you learned — with my publications on a timeline, the paper-to-valuation loop, and every claim linked to its source.
Invention & IP · 29 September 2026 · 7 min read · Read the note →
From LSA to LLMs: twenty years of teaching machines meaning
In 2004 I built LSAbot, a chatbot that chose its answers by geometry rather than by keywords. What carried over into today’s language models, what did not, and why I still keep one foot in formal methods — with my publications and certifications on a timeline.
Research · 6 October 2026 · 8 min read · Read the note →
