Making AI stick
Models are probabilistic. Products have to be deterministic. The interesting work is the layer in between: the harnesses, evals and guardrails that decide where a guess is acceptable and where it never is.
Prototypes, the frameworks behind them, books worth arguing with, and parenting in the AI era. Plus coffee, and the occasional complaint.
Lately
- Rain in NovemberA letter to Kiddo, written the week after our first rock concert.
- Enterprise AI EvalsWrite your evals as a PRD, so the team agrees on what good looks like before arguing about it.
- To (AI) Engineer Is HumanBlack October, a sandbox escape, and why only a swarm can defend against a swarm.
- Site2VoiceThe moving parts of a voice AI agent, worked through on one that turns any website into a sales rep. Prototype.
- The Singularity Is NearerNanobots, longer lives, and a basic income that doesn't travel well.
- Kaise Chal Raha Hai?A new footpath, three months, and a rulebook nobody walks.
- 7 PowersWhy some companies keep winning long after everyone has seen how.
- Context PassportYour context, owned by you, handed to each AI tool in slices. Concept.