Power Rankings

As co-founder and CEO of LangChain, Chase is a central figure in LLM agents and memory systems. LangChain is the de facto standard agent framework (for Python and JS) that first connected LLMs with tools and actions. Under Chase's guidance, LangChain has introduced high-level memory modules (vector memories, summarization memories, etc.) and recently launched LangGraph for managing agent context. His Sequoia-backed platform (valued over $1B) means he shapes much of the tooling and best practices around RAG and agent memory.

Liu is co-founder/CEO of LlamaIndex (formerly GPT-Index), a popular data framework for LLM applications. LlamaIndex provides structured data ingestion and vector-based retrieval (RAG) tools that effectively serve as external memory for LLMs. Under Liu's leadership, LlamaIndex became widely adopted for building LLM knowledge graphs and memory buffers, cementing his influence on memory architectures (like semantic, document, and graph memory) in the agentic AI community.

The creator of AutoGPT, Richards (aka "Significant Gravitas") ignited mainstream interest in autonomous LLM agents. AutoGPT is an open-source agent that automatically breaks goals into sub-tasks and manages its own short-term memory to complete long-term objectives. Although not an academic, Richards's early 2023 release of AutoGPT (on GitHub) and its viral demos made him a highly visible innovator. His project demonstrated both the power and the limitations of LLM agents' memory, spurring efforts to add better memory modules.

A senior researcher at MSR AI Frontiers, Awadallah co-led AutoGen, Microsoft's multi-agent LLM framework. AutoGen lets developers compose teams of LLM-based agents with shared memory and knowledge (including "teachable" memory updates). The published AutoGen paper and tools (COLM 2024, open-source GitHub) are widely cited by industry as enabling agents with robust memory modules, making Awadallah a key figure in agentic memory systems.

An AI researcher at Rutgers and the AIOS Foundation, Xu introduced A-Mem: Agentic Memory for LLM agents. His work applies the Zettelkasten linking method to LLM memory: new "memories" automatically get structured notes and links to related past memories, with dynamic evolution of context vectors. By formalizing agent-driven memory updating, Xu's A-Mem paper is among the first systems explicitly targeting long-term memory organization for LLM agents.

A PhD alumnus of Berkeley and now co-founder/CEO of startup Letta, Packer was the lead author of MemGPT. He originated the idea of teaching an LLM to manage "main memory" vs. "external memory" much like an OS. His open-source MemGPT code and talks have popularized the OS-inspired memory model for LLM agents. This earned him visibility as a technical innovator in LLM memory systems.

A prominent computer systems researcher, Prof. Stoica co-led development of MemGPT, an OS-inspired memory-management framework for LLMs. As co-author on the MemGPT paper, he helped design a hierarchical memory architecture (main vs. external memory) that lets LLM agents "swap" information to extend context. Stoica's leadership in scaling AI systems and his high profile at Berkeley AI lend significant clout to the agentic-memory field.

A researcher specializing in memory and language, Toneva co-authored recent position papers on episodic memory in LLM agents. Formerly a senior scientist at Meta/Facebook AI, she now works at MPI-SWS and has championed modeling how LLMs can store and retrieve episodic details. Her work on integrating long-term, context-specific memory (drawing on cognitive memory theory) makes her a key academic influencer in agentic memory research.

A cognitive neuroscientist who helped pioneer computational memory models, Dr. Norman co-authored recent work evaluating episodic memory in LLMs. His Princeton lab's models of human memory inform memory architectures for AI, and he is a lead author on an arXiv paper on LLMs' sequence-recall tasks (episodic memory). Norman's insights into how agents remember context and temporal order position him as a thought leader on memory fundamentals in autonomous LLM systems.