985: The Four Types of Memory Every AI Agent Needs, with Richmond Alake
Get the full intelligence
Search transcripts, export clips, track mentions, and explore all topics from “985: The Four Types of Memory Every AI Agent Needs, with Richmond Alake” inside PodZeus.
In this episode of the Super Data Science Podcast, host Jon Krohn welcomes back Richmond Alake, Director of AI Developer Experience at Oracle, to explore the critical yet often overlooked component of AI agent development: agent memory. Alake defines agent memory as the encapsulation of systems—such as LLMs, databases, and rerankers—that enable AI agents to learn, adapt, and retain context over time. He breaks down the four types of agent memory inspired by human cognition: episodic (time-stamped experiences), semantic (factual knowledge), procedural (skills and routines), and working memory (real-time context). The conversation highlights why memory is foundational to building effective, production-ready agents, especially as RAG (Retrieval-Augmented Generation) alone falls short in handling memory updates, conflicts, and forgetting. Alake emphasizes the emerging role of 'memory engineers' and advocates for a 'memory-first' agent harness approach to reduce cognitive load by consolidating data types—vector, graph, relational, and spatial—into a unified system like Oracle's AI Database. He also discusses the importance of reducing fragmentation in the agent stack, the value of open-source tools like AgentSpec, and the need for developers to embrace lifelong learning in this fast-moving field. The episode concludes with practical resources, including a Deep Learning AI course on memory-aware agents and upcoming O'Reilly boot camps.
Agent memory is essential for AI agents to maintain continuity, adapt over time, and deliver real business value.
There are four types of agent memory: episodic (time-based experiences), semantic (factual knowledge), procedural (skills and routines), and working memory (real-time context).
RAG is insufficient for agents because it doesn’t handle memory updates, conflict resolution, or forgetting—key functions of true memory.
A 'memory-first' agent harness reduces cognitive load by using a unified database (like Oracle AI Database) to manage multiple data types.
Developers should focus on domain-specific workflows and leverage open-source tools and courses to accelerate learning in agent memory.
…and 1 more takeaway available in PodZeus
The Critical Role of Agent Memory in AI Development
“Memory is one of the critical components of building AI agents that actually work in production and actually create value.”
Defining the Four Types of Agent Memory
“Working memory is what you're using in real time, in the context. That's my working memory. The best way I would describe this within a genetic context is the context window of the LLM.”
Beyond RAG: Why Memory Engineering Is Essential
“RAG doesn't cover all of that. You need to update memory, consolidate memory, resolve conflicts, and forget memory.”
The Rise of the Memory Engineer and the Age of Memory
Alake discusses how he coined the term 'memory engineer' to describe professionals bridging database expertise with AI agent development. He reflects on his '100 Days of Agent Memory' initiative, which helped educate developers and validate the growing importance of memory in AI.
Building a Memory-First Agent Harness
“If you just take that mindset shift—thinking the information that comes into this system needs to be recalled and forgotten—it just changes the way you approach every aspect of your data modeling.”
“If you just take that mindset shift—thinking the information that comes into this system needs to be recalled and forgotten—it just changes the way you approach every aspect of your data modeling.”
“Memory is one of the critical components of building AI agents that actually work in production and actually create value.”
“RAG doesn't cover all of that. You need to update memory, consolidate memory, resolve conflicts, and forget memory.”
Host
Guest
Oracle
organization
Richmond Alake
person
Jon Krohn
person
Deep Learning AI
organization
Andrew Ng
person
O'Reilly
organization
Hubel and Wiesel
person
Anthropic
organization
Excel Data
organization
Fei-Fei Li
person
979: Agentic Data Management and the Future of Enterprise AI, with Rohit Choudhary
Super Data Science: ML & AI Podcast with Jon Krohn • 1h 5m • 3/31/2026
981: How Data Engineers Are “10x’ing” Themselves With Agents, feat. Matt Glickman
Super Data Science: ML & AI Podcast with Jon Krohn • 1h 14m • 4/7/2026
982: In Case You Missed It in March 2026
Super Data Science: ML & AI Podcast with Jon Krohn • 44m • 4/10/2026
983: AI in the Classroom: How a Top Elementary School Is Doing It Right, with Principal Traci Walker Griffith
Super Data Science: ML & AI Podcast with Jon Krohn • 1h 12m • 4/14/2026
984: Building AI Agents Where 99.9% Accuracy Isn't Good Enough, with Raju Malhotra
Super Data Science: ML & AI Podcast with Jon Krohn • 29m • 4/17/2026
Get the full intelligence
Search transcripts, export clips, track mentions, and explore all topics from “985: The Four Types of Memory Every AI Agent Needs, with Richmond Alake” inside PodZeus.
Start discovering podcast insights today
Start with a 7-day trial and explore a growing catalog of popular podcasts. No credit card required.
No credit card required • 7-day trial • Cancel anytime