▶ Blog
Thoughts on AI systems, prompt engineering, safety, and what it takes to build reliable AI in production.
WeftProgramming LanguagesAI SystemsOpen Source
The Future of Programming (and Why I'm Building a New Language)
Programming changed in two ways at once: an AI writes the code now, and programs are made of models and people. The languages we use were designed for neither, so I am building one that is.
October 4, 2026 · 5 min read
Read
AlignmentAI Safety
Three Properties for Alignment (and Why We're Not Training Them)
Alignment means nothing without specifying what you're aligned to. I propose three concrete, mechanistic properties (Red Lines, Embodiment and Resilience) and sketch a training procedure that could get us there.
March 16, 2026 · 12 min read
Read
Prompt EngineeringLLMs
The Topology of LLM Behavior
A mental model for prompt engineering: think of LLM generation as a weighted random walk through a semantic space shaped by attractors. Two distinct operations follow from this: navigating around the attractors, and reshaping the topology itself.
February 27, 2026 · 9 min read
Read