Memory in the Age of AI Agents (Dec 2025)

Abstract

• Title: Memory in the Age of AI Agents (Dec 2025)
• Link: http://arxiv.org/abs/2512.13564v1
• Date: December 2025

Summary:

This survey provides a comprehensive landscape of memory systems for LLM-based agents, proposing a unified taxonomy based on Forms (Token-level, Parametric, Latent), Functions (Factual, Experiential, Working), and Dynamics (Formation, Evolution, Retrieval). It distinguishes agent memory from related concepts like RAG and context engineering, systematically reviews existing architectures and benchmarks, and identifies emerging frontiers such as generative memory, reinforcement learning integration, and self-optimizing memory management.

Key Topics:

• Agent Memory Taxonomy
• Memory Forms (Token-level, Parametric, Latent)
• Memory Functions (Factual, Experiential, Working)
• Memory Dynamics (Formation, Evolution, Retrieval)
• Generative Memory
• Reinforcement Learning Integration
• Multimodal Memory
• Trustworthy Memory