Remember me, memory in agents | The Agent Factory Podcast

How do you build AI agents that actually remember and learn from past conversations? This episode of The Agent Factory is a masterclass on agent memory, a critical component for building smarter, more personalized agents. Hosts Ivan Nardini and Annie Wang, along with guest Kimberly Milam, dive into the challenges of short-term memory (sessions) and introduce a powerful new solution: Vertex AI Memory Bank, a managed service designed to give agents persistent, human-like long-term memory. 🧠 By the end of this episode, you'll have a deep understanding of why memory is essential for agents, and how to implement it to create more capable and engaging AI. In this episode you will learn: 1️⃣ Agent Industry Pulse: We'll share the latest and greatest updates in the world of AI agents. We explore new models like Zhipu AI's GLM-4.5, which are optimized for agentic tasks. We also cover new evaluation frameworks like LangChain's Align Evals and the latest features in the Agent Development Kit (ADK) and the agent to agent protocol. 2️⃣ The Factory Floor: The core of this episode is a deep dive into memory. We'll start with the basics, exploring the difference between short term memory (sessions) and long term memory. We'll show you why a "digital goldfish" agent fails to remember past interactions and how a session based approach solves this for a single conversation. Then, we'll introduce the solution for long term memory: Vertex AI Memory Bank. 3️⃣ Deep Dive on Vertex AI Memory Bank: Our guest, Kimberly Milam, a core contributor to Memory Bank, explains the service in detail. 4️⃣ The core difference between Memory Bank and RAG: How Memory Bank uses LLMs to agentically manage memory, deciding what to extract, update, or delete, as opposed to vanilla RAG which uses static corpuses. 5️⃣ How Memory Bank works: The two main processes of memory extraction and memory consolidation that ensure memories are meaningful, de-duplicated, and up to date. 6️⃣ How to integrate Memory Bank: