About the session
For two decades, personalization has largely meant ranking content, products, or actions based on past behaviour. Agentic AI creates a more ambitious possibility: systems that understand a user’s intent, reason across short- and long-term preferences, ask clarifying questions, use tools, and help the user accomplish a goal.
But turning a recommender system into a trusted personal agent is not simply an LLM integration problem. It requires a new architecture for user understanding, memory, planning, retrieval, recommendation, evaluation, and human control.
Drawing on lessons from building personalization and user-intelligence systems at large scale, this talk presents a practical blueprint for the agentic personalization layer. It explores what should be learned from behavioural data, what should be explicitly confirmed with users, how short-term intent should interact with long-term preferences, and where human agency must override algorithmic confidence.
The talk will also examine the product and organizational changes required to move from optimizing clicks to helping people make better decisions."
