Most AI initiatives start with promising LLM demos, but turning them into reliable, scalable AI agents in production is a different challenge altogether. In this session, Philips Hue shares its end-to-end journey from early experimentation to operating AI agents at scale. We’ll explore how we evolved from monolithic LLM applications to modular, task-oriented agent architectures, and how this shift enabled better maintainability, observability, and control. The session covers key design decisions around orchestration, tool integration. You’ll gain real-world insights into what works, what doesn’t, and what it truly takes to run AI agents in production environments.