Answers are grounded in this paper and cite their sources.
Recent position papers and surveys have identified episodic memory as thecritical missing component in large language model (LLM) agents. Whileepisodic memory is necessary, I argue it is far from sufficient. Drawing on fivedecades of cognitive science research and a systematic analysis of contemporaryagentic memory systems, this paper proposes a unified memory lifecycle modelbuilt on six interdependent cognitive pillars: Encoding, Storage andOrganization, Retrieval, Consolidation, Forgetting, and Metamemory. Each ofthese mechanisms has been studied in isolation across cognitive science,neuroscience, and machine learning. The contribution here is integrating theminto a single lifecycle framework and arguing that the neglected pillars, not thewell studied ones, are what separate memory augmented agents from genuinelymemory competent ones. Through a parallel examination of how biologyhandles each pillar and how current agentic systems attempt to replicate it, Iidentify a persistent asymmetry: the research community has invested heavily inhow to put memories in and get them out, while neglecting the mechanisms thatgive biological memory its adaptive power.
No takes yet. Share an insight, caveat, or question.
Sandhanakrishnan Ravichandran (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: