Industrial systems generate vast amounts of heterogeneous data — but turning that data into actionable knowledge remains a hard problem. This session explores how agentic AI systems, powered by knowledge graphs and large language models, can reason over complex industrial environments to support decision-making, predictive maintenance, and process optimization. We'll look at where current RAG and LLM architectures genuinely add value in manufacturing contexts — and where they fall short. Drawing on real-world industrial use cases, the talk bridges semantic technologies, AI system design, and operational constraints. Expect honest engineering tradeoffs, not hype.