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'll Learn:
- 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.
- 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.
- Deep Dive on Vertex AI Memory Bank: Our guest, Kimberly Milam, a core contributor to Memory Bank, explains the service in detail. You'll learn:
- 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.
- 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.
- How to integrate Memory Bank: A simple guide on how to integrate the service with agent frameworks like ADK, LangGraph, and CrewAI using a clear API.
Sessions vs. Memories: A clear breakdown of the difference between a session (conversation history) and a memory (a self-contained, meaningful fact).
- Management and Scope: We discuss how memories are self-managed and how using a scope (like a team_id or user_id) allows for shared knowledge bases and multi-user access.
- Developer Q&A: We answer top community questions on building with agents, including:
What are the official online communities for ADK?
How can you make an ADK agent act deterministically?
How can you build a collaborative agent where the Memory Bank is shared across a team?
About The Agent Factory:
"The Agent Factory" is a video-first technical podcast for developers, by developers, focused on building production-ready AI agents. We'll explore how to design, build, deploy, and manage agents that bring real value using the latest open-source tools and frameworks. Going beyond basic AI/ML concepts, the podcast will delve into the practical aspects of agentic AI, covering everything from single-purpose agents for task automation to complex multi-agent systems.
Connect with Us & Ask Questions:
- Leave your questions and feedback in the YouTube comments below!
- Join the conversation on social media with the hashtag #TheAgentFactory.
- Connect with the community at the Google Developer Program forums → http://goo.gle/4mH1gFY
Resources:
Agent Industry Pulse:
- LangSmith Align Eval feature → http://goo.gle/45pZmE5
- Agent-to-Agent protocol blog post → http://goo.gle/4oI4HhA
- Zhipu AI blog post → http://goo.gle/4lISaYE
Memory Bank:
- VertexAI Official documentation → http://goo.gle/4mLlFto
- Vertex AI Express mode Official documentation → http://goo.gle/4mr2c1y
- Get started with Memory bank using ADK → http://goo.gle/4mlwkLE
- Get started with Memory bank using LangGraph → http://goo.gle/41LNfyL
- Get started with Memory bank using CrewAI → http://goo.gle/4mwAdgX
- Official blog post → http://goo.gle/4mUjggp
- Developer blog post → http://goo.gle/46YjyOx
- Agent engine lab → http://goo.gle/41Jnejz
- Agent tutorial lab → http://goo.gle/4fDNS36
Developer Q&A:
- Agent Development Kit GitHub Discussions → http://goo.gle/3HFut5g
- ADK Subreddit → http://goo.gle/4fI4Hd7
- Manually transfer to agent → http://goo.gle/4lxxpia
- Generate memories → http://goo.gle/45nGDc7
#VertexAI #ADK #GoogleCloud
Speaker: Annie Wang Ivan Nardini Kimberly Milam
Products Mentioned: Vertex AI, LangChain, ADK