arxiv.org/abs/2610.34073 been sayin this; straightforward naive RAG search over a corpus reliably outperforms complex memory systems, with summarization, fact extraction, dreaming etc mostly hurting rather than improving effectiveness
Memory Tax: Dissecting the Net Value of Agent Memory
Memory is increasingly used in long-horizon agents, yet the net value of building memory over direct retrieval from raw history remains largely unclear. Through a large-scale evaluation across multipl...
arxiv.org
