Architecting Agent Memory: Principles, Patterns, and Best Practices — Richmond Alake, MongoDB

Richmond Alake delivers a presentation about the importance of memory in AI agents and agentic systems, emphasizing how memory enables agents to be more reflective, interactive, proactive, reactive, and autonomous. He discusses the evolution from chatbots to AI agents, highlighting the different forms of memory such as short-term, long-term, conversational, and workflow memory. Alake introduces MongoDB as a memory provider for agentic systems, offering features to turn data into memory and improve the believability, capability, and reliability of AI agents. He also mentions the open-source library Memoriz, which encapsulates memory implementation and design patterns, and the acquisition of Voyage AI by MongoDB to enhance embedding models and reduce AI hallucination. The presentation concludes with the importance of drawing inspiration from neuroscience and nature to build effective agentic systems.