What I'm doing now
Current focus: Building distribution retrieval and ranking infrastructure that serves billions of queries. Day to day, I think about search quality, latency budgets, cache economics, and the gap between offline benchmarks and what users actually experience.
Building: The Against Entropy site you're reading. I'm restructuring it into a focused publication on AI systems, retrieval, agents, and product strategy. Essays every couple of weeks, short notes in between.
Learning: Multi-agent orchestration patterns, eval design for RAG systems, and the economics of caching in AI inference. Slowly working through the Index Zero course material I built — refining it as I learn more.
Open questions I'm chewing on:
- How do you build evals that don't degrade the moment your data distribution shifts?
- What's the right boundary between a skill (deterministic tool) and an agent (autonomous decision)?
- Why does every caching discussion in AI start with the model and never with the query path?
Recently changed my mind about: I used to think agent frameworks were mostly over-engineering. I still think half of them are, but the orchestration problem at scale is real and the patterns emerging are genuinely interesting.
Last updated May 2026. Inspired by Derek Sivers' Now page movement.