PolyAI on building context-aware voice agents: latency, knowledge bases, and what actually ships

In a discussion between SurrealDB CEO Tobie Morgan-Hitchcock and PolyAI CTO Shawn Wen, they explore the challenges of deploying context-aware AI agents in enterprise contact centers, particularly for voice interactions where latency and accuracy are critical. The conversation highlights that while technology like RAG (Retrieval-Augmented Generation) isn't the differentiator, engineering disciplines around context, observability, and data ownership are vital. They stress the importance of treating the knowledge base like code, with versioning and governance, to ensure reliable and repeatable outputs. The discussion also underscores the need for enterprises to validate outcomes before heavily investing in centralizing knowledge bases or training their models. PolyAI's upcoming developer platform aims to facilitate the integration of voice and chat agents into existing enterprise systems, emphasizing model-agnostic and integration-first design principles. The session serves as a reminder that in customer support, especially with voice, latency and consistency are more than metrics—they define the user experience.