Beyond Chatbots: Agentic Systems as Cybernetic Control Loops

AI agents are usually presented as chatbots with tools. That framing is too small. If agents are going to matter in the real world, they need to reason over domain models, respect constraints, observe outcomes, and improve from feedback. In other words, they need ontology, operations research, and data systems that preserve ground truth. This talk uses advanced manufacturing as the concrete setting, because a factory is a harsh test for AI: decisions touch machines, people, materials, schedules, quality records, and customer commitments, and correctness cannot be faked with a better prompt. The system needs a clean model of the world, traceable data, immutable history, human correction loops, optimization engines, and observability from raw unit-level events up through business outcomes. It walks through a practical architecture for agentic systems that combine LLMs, optimization, simulation, and data engineering, covering where agents are useful, where they are dangerous, and why the next generation of AI systems will look less like copilots and more like cybernetic control systems: sensing, deciding, acting, measuring, and learning. The goal is not to sell hype. It is to give data scientists, data engineers, and technical leaders a sharper mental model for building AI systems that can survive contact with reality.