Feeding and Fencing AI Agents: Engineering Governed Agentic Systems on Real-Time Data
This session explores how to make autonomous AI agents trustworthy at scale. Using a fraud investigation scenario built on Apache Kafka, the Model Context Protocol (MCP), and an LLM-based reasoning agent, the talk shows how governance can be engineered into both the real-time data layer and the tool boundary that controls agent actions. Attendees will learn - Why the trust boundary in an agentic system belongs in the server, not only in the prompt - How MCP can be used as a typed, auditable governance contract - Engineering patterns behind shadow mode, approval-gated tools, and multi-model routing - How these concepts come together in a working open-source reference implementation they can explore afterward This session is designed for engineers, architects, product leaders, researchers, students, and anyone interested in building trustworthy AI-agent systems. Co-sponsored by: Central Indiana IEEE Computer Society Virtual: https://events.vtools.ieee.org/m/563554