Recent advances in large language models (LLMs) and agent-based AI systems have enabled intelligent applications that interact with complex data, environments, and users. Such systems are increasingly deployed in application scenarios including decision support, scientific discovery, personalized education, software engineering, and interactive data analysis, where agents are expected to operate over extended time horizons, maintain contextual awareness, adapt to new information, and improve performance through experience. Achieving these capabilities requires principled approaches to memory management, knowledge updating, and agent evolution. These areas remain open and actively evolving research challenges.

This Research Topic aims to bring together recent advances and emerging perspectives on memory, knowledge updating, and evolution in AI agents, with a focus on scalable methods, reliable learning dynamics, and practical deployment in data-intensive settings. We seek contributions that investigate how agents can accumulate, organize, and retrieve long-term experience; update internal knowledge in a controlled, efficient, and reliable manner; and adapt reasoning strategies or behaviors in response to changing tasks, environments, or objectives. These challenges are closely connected to broader issues of efficiency, robustness, alignment, and continual learning in modern AI systems.

The scope of this Research Topic includes, but is not limited to, the following research directions:

●LLM / Agent (Multimodal) Memory

●Knowledge Editing & LLM / Agent Steering

●Efficient LLM Reasoning

●Agent Evolution (long-term capability development and adaptation through interaction or learning)

●LLM / Agent Innovation (the emergence of novel strategies, representations, or problem-solving behaviors beyond predefined routines)

In addition to algorithmic and system-level contributions, this Research Topic explicitly encourages work on evaluation methodologies and benchmarking for memory-enabled and evolving agents. Relevant submissions may propose new benchmarks, datasets, experimental protocols, or evaluation metrics that assess long-term adaptation, memory utilization, knowledge consistency, reasoning efficiency, or agent robustness across tasks and time.

We welcome original research articles, methodological contributions, system and architecture designs, benchmark datasets, and empirical studies addressing one or more of the above directions. Submissions may include theoretical analyses, experimental investigations, or applied studies demonstrating memory-aware, adaptive, or evolving agents in real-world or large-scale data scenarios.

By consolidating research across these interconnected themes, this Research Topic aims to advance the understanding of how AI agents can support long-term adaptation, reliable knowledge management, and efficient reasoning, and to provide a shared forum for researchers and practitioners working on next-generation AI agents capable of sustained operation in dynamic and data-rich environments.

Are you interested in contributing a manuscript to this Research Topic?

Memory, Knowledge Updating, and Evolution in AI Agents