Notes & ideas
Thinking out loud.
Experiments, questions, and things worth writing down.
Evaluation Inside the Agent Loop
A prediction before each action can turn an ordinary execution trace into a stream of graded claims.
Agent evaluation · 3 min readAI Text Watermarks in Practice
A closer look at statistical watermarks, detection, and the practical limits of rewriting AI-generated text.
AI & governance · 3 min readWhen Client Data Has to Stay Local
What a 100,000-document archive taught me about local models and the constraints that shape AI adoption.
Local AI · 1 min read