Why LLM Agent Memory Fails Under Continuous Updates

A visual explainer of arXiv:2605.12978 showing how repeated LLM memory consolidation can turn useful experience into faulty rules—and why agent systems should retain raw episodes, gate abstraction, and preserve traceable evidence.