Eval-Driven Development
AI systems don’t fail because they’re dumb, they fail because we never define “good.”
Jack Arenas
Search
All Posts
Adapting Foundation Models: From Prompts to Reinforcement Learning
The three levers for steering pre-trained models and the evals that guide them.
When Chat Replaces Search: The Economics of AI-Mediated Information
When answers replace clicks, publishers, brands, and platforms face a new incentive structure.
The Promptware Window
Every week I see pitches that follow a similar pattern: “X for Y, powered by AI.” I call this promptware.
The AI-Native CTO
Why technical leadership just got exponentially harder.
Human Ingenuity as the Scaling Limit
Apple’s Illusion of Thinking reminds us that transformers may asymptote to human ability and that may be enough to transform the world.