Building Production-Ready Multi-Agent Systems

Multi-agent systems promise to split complex work between specialised agents — but most prototypes fall apart under real traffic. This keynote walks through the architecture patterns that held up in production: supervisor/worker topologies, explicit hand-off contracts, and evaluation loops that catch regressions before users do.