Field notes with receipts
Each entry is a documented experiment: what was built or decided, what the evidence actually showed, and what happened next. Not arguments. Proof.
What happens when you build a decision system that runs technically but never closes a real consequence contract?
157 sprints. A live system. A bake-off it won. A consequence contract still open. What building a capital decision system actually proved.
What can AI-era decision tools reveal about a stay-or-leave startup call that was originally made on instinct?
I exited a data privacy startup in 2022 on gut and friction. Four years later, AI-era tools let me revisit that decision with a structured thinking protocol and a probability model. The real finding: the gut had already computed the answer. The model gave me language for why.
What happens when AI lets you ship something before you know whether you should trust it?
A field note on the trust gap between AI-enabled output and the judgment required to ship, own, and rely on it.
What has actually been proven when an AI-built system goes live?
Why a system can be live, healthy, and still not be understood or useful.
What coordination does cheap AI execution still require?
A field report on Relayfox, file-native coordination, and the coordination tax that remains when smaller models do bounded AI-assisted work.