Presentation: From Models to Agents: Building Context-Aware Consumer AI at Scale at DoorDash

DoorDash is transitioning its recommendation systems from basic predictive models to agentic platforms capable of maintaining consumer memory and context. By utilizing semantic IDs and grounded search to represent its catalog, the company aims to improve recommendation relevance and increase conversion rates. This shift reflects a broader industry trend toward AI agents that provide personalized, stateful experiences rather than executing isolated, single-step tasks.