Whether they’re part of a workflow or powering a user experience, AI agents often need to work on your data at interactive speeds. Because of this, they demand more from data systems than traditional analytics workloads. Low latency is no longer a luxury, and handling concurrency is essential if you want an agent to scale with a growing platform. But with great access comes great risk: safeguards and strict permission models are critical. This talk will explore the principles and patterns for building responsive, reliable data infrastructure that agents can trust and users can depend on.