Is Complex Agentic Memory Always Needed? Naive RAG Is Stronger Than We Thought

[arXiv] Zhengye Bao, Shengtao Zhang and others study Raw-RAG: dense retrieval and reranking over raw dialogue history, which is then supplied to a reader large model to answer. The authors claim that on LoCoMo, LongMemEval, MEME and MemoryAgentBench, Raw-RAG ranks first in six of seven benchmark-model settings, scoring 0.13 to 5.53 points higher than the strongest compared agent memory system. The paper claims that higher evidence coverage can offset lower utilization; even with full access to a transformed memory store, an accuracy gap relative to raw evidence remains, challenging the necessity of complex agent memory on the evaluated dialogue tasks.