Designing Scalable Agent-Based Service Architectures

Description

With more agentic workflows moving into production, businesses must rethink not just application logic but also the underlying database layer and its scalability. Rob Reid, Technical Evangelist at Cockroach Labs, examines an agentic fraud detection system to explore what AI-ready scale looks like in practice. You’ll see how three coordinated agents—anomaly detection, reasoning, and notifications—operate continuously under load, use vector embeddings and L2 distance to identify anomalous transactions, fetch real-time context, reason about risk, and trigger customer notifications. You’ll also learn about the database architecture this model needs in order to handle high concurrency, maintain strong consistency, and unify relational and vector data without brittle orchestration or batch pipelines. Recorded as part of the O'Reilly Software Architecture Superstream: Software Architecture and the Age of Agentic AI on March 5, 2026.