Writing from inside the AI wave
I write about working with AI agents as they actually behave, not as advertised. Newest first.
-
The Comment That Lied
Four of sixteen agent memories had gone stale — invalidated by the fixes we'd just shipped. Institutional memory rots, and progress is what rots it.
-
Scheme in a Weekend, or, LLM: The Ultimate Intern
One weekend, 13,000 lines of Rust, all 1,225 conformance tests passing — and what it taught me about managing an AI that keeps declaring “done”.
-
The Persistence of Memory
Agent memory files are configuration, not notes — and for the first time, the configuration writes itself.
-
The Slash Command That Knew Too Much
My development workflows are now programs written in English, run by an LLM.
-
The Joy of Rediscovery
An earlier piece, from the first months of ChatGPT: I love tools that unearth the forgotten past — and then a new one arrived that could read the whole library back to me.
These also appear on Medium.