Summary

Sandy, an expert in distributed data systems, shares insights on the complexities of scaling multi-agent AI systems from one to multiple agents in production environments. Drawing from her experience at AWS and Databricks, she highlights how adding agents transforms an AI project into a distributed systems challenge, with intricate coordination, state management, and failure recovery requirements. Her key message is that expanding AI agent systems is not simply about adding features, but requires careful architectural design to manage interdependencies, race conditions, and potential workflow breakdowns across interconnected agents.