Blog
Writing
Notes on AI engineering, the books I'm reading, and the occasional detour into philosophy and psychology. RSS
Notes on building LLM agents that don't fall over
Agents fail in boring, predictable ways: unbounded loops, malformed tool calls, and silent drift. A few patterns that make them sturdier, with a minimal Python loop.
Stoicism and the art of debugging
The dichotomy of control, the view from above, and premeditatio malorum — three Stoic ideas that turned out to be excellent debugging tools.
Thinking, Fast and Slow, read by an engineer
Kahneman's System 1 and System 2 map uncomfortably well onto how we build, evaluate and trust machine learning systems — and onto how I make decisions at work.
Why I'm starting this blog
Writing is how I find out what I actually think. Here's what this blog is for, what I'll write about, and what I won't pretend to know.