May 20269 min read
→The Control Boundary for Agent Systems
Why agent safety belongs at the action boundary: tools, databases, APIs, and side effects.
Lab Notes
engineering notes and thinking from membrane labs.
Engineering notes on agent policy, retrieval, memory, routing, and the runtime boundaries where agent systems become real software.
April 202618 min read
→Your RAG Cache Is Serving Wrong Answers
A correctness failure mode in semantic caching for RAG systems, with a retrieval-aware validation layer.
April 202614 min read
→Access Patterns for Caching in AI Agent and RAG Systems
A taxonomy of access patterns across enterprise RAG and agentic deployments, and what each requires from a cache.