Blog
Ideas, opened carefully.
The interesting problems begin when the assumptions stop holding.
Headless Is an Architecture, Not a Product
Headless architecture and AI agents can remove interface friction. They do not automatically solve trust, security, quality, governance, or cost.
A Benchmark Is a Game Against the Model
Gemini 3.8 Flash's Terminal-Bench split shows how public LLM benchmarks become strategic games between model developers and evaluators.
The Price of Seeing
How safe exploration can create the evidence needed to learn hidden behavior in imperfect-information games.
The Gap to AGI: Learning How to Think
General intelligence may require adapting the form of computation, not only scaling the amount of reasoning.