Memory-Aware Agent Harnesses for Anti-Money Laundering — Nacho Martinez, Oracle

This post details the application of agent memory and harness engineering to an Anti-Money Laundering (AML) autonomous agent. It elaborates on the agent stack, including short-term, episodic, procedural, and semantic memory, and discusses context summarization vs. compaction. The agent loop, toolbox and skillbox patterns, skill distillation, and continual learning for improving frozen models are explained. The talk also covers instruction hierarchy and provides a UI walkthrough. Q&A sessions address episodic memory conflicts, skills, caching, context budgets, memory layer simplicity, pre-built SQL vs. dynamic queries, and hybrid search.