From Context to Memory: Your Agents Need a Real Memory Layer — Anders Swanson, Oracle

Why watch

Anders Swanson from Oracle demonstrates why context windows and scratch files fail as memory solutions for AI agents, and presents a comprehensive memory layer architecture. The talk covers typed memory patterns, storage schemas with vectors/text/graph/audit history, safe memory formation with redaction, hybrid search retrieval, and handling failure modes like stale, poisoned, and conflicting memories. Ideal for developers building AI agents who need persistent, recallable memory systems.