Scaling AI Agents in Practice: Patterns, Quality, and Ownership in a Multi Agent Product

Most teams can ship an AI agent. The harder part is scaling it across multiple squads, product lines and customer contexts without ending up with inconsistent experiences and unreliable outputs. In this forum, Deb Kawamoto shares how Vanta moved from a fast “ship and learn” agent launch to shaping a multi-agent experience that can scale. She will outline the principles and early interaction patterns her teams used to create consistency, plus the operational changes required once adoption and scope expanded. The discussion then opens to the audience to dig into what breaks in practice: fragmented quality validation, unclear ownership across functions, and the reality that data structure and system constraints can make or break the experience.

What you will learn:

• Starting principles and UX patterns that help multiple teams ship agent experiences without fragmentation.
• Practical approaches to validating response quality when teams move fast and methods differ.
• How to define ownership across design, product, engineering, and data when AI outcomes depend on all of them.
• What needs to change in ways of working as AI moves from experiment to scaled product capability.