Agent Memory as Core Design: From Pilot to Production - Varun Yadavalli

This talk focuses on the transition of AI agents from pilot projects to production environments, addressing the critical challenge of memory in enhancing agent performance. The speaker emphasizes that memory is a fundamental aspect of production-grade AI systems, positing that without it, agents struggle to retain learning between sessions, leading teams to rely on cumbersome prompts and repeated instructions. Through a document extraction use case, the session illustrates how memory allows agents to maintain useful context, adapt across different runs, and continuously improve. The presentation also delves into the various types of memory and provides insights on how to effectively design memory systems that foster adaptation, learning, and maintainability.