Jul 20, 2026
Mark Nelson explains how agent memory helps AI systems retain
useful context, balance different memory types, and support secure,
production-ready AI applications. The discussion explores why
governed, durable memory and Oracle AI Database provide a strong
foundation for enterprise AI systems.
In this episode of Digital Impact Radio Series
9, Franco Ucci speaks with Mark Nelson about one of the
most important concepts in modern AI development: agent memory.
Mark explains why agent memory is much more than chat history and
how developers should think about memory as structured, reusable
application data rather than a simple transcript of previous
conversations.
The discussion introduces six core memory types, including working,
semantic, episodic, procedural, relationship, and operational
memory, and explains the role each plays in building intelligent AI
agents. Mark also examines how popular AI frameworks approach
memory, why developers should begin with a clear memory policy
instead of a storage technology, and how retrieval, governance,
privacy, explainability, and lifecycle management all contribute to
trustworthy enterprise AI.
The conversation concludes with a practical discussion of Oracle
Agent Memory and Oracle AI Database, showing how vectors,
relational data, metadata, graph relationships, and enterprise
governance can work together to create durable, searchable,
well-scoped, and auditable memory for AI agents. Whether you are
building AI applications, experimenting with agent frameworks, or
designing enterprise AI architectures, this episode provides a
practical framework for understanding how memory enables more
capable and reliable AI systems.