How I built an intent drift detector for LLM agents

This article details the creation of SIP (State Integrity Protocol), a tool designed to detect intent and semantic drift in LLM agent outputs. It addresses the silent failure problem of AI agents by automatically checking for discrepancies between expected and actual outcomes.