Field notes on reliable AI.
Practical writing on evaluation, drift, cost, and the engineering that keeps AI systems working after launch.
How to Tell Whether Your AI System Is Drifting
Four kinds of drift degrade AI systems: model drift, data drift, LLM behavioral change, and silent vendor updates. Here is how to catch all four.
AI Memory Is Architecture, Not Magic
AI memory is not a model feature. It is a system you design: short-term context, long-term storage, retrieval, expiration, isolation, and a real cost curve.
The Model May Not Be Your AI Problem
Most production AI failures blamed on the model come from retrieval, data quality, context assembly, or workflow logic. Here is how to find the real cause.