Security teams don't struggle because they lack data—they struggle because connecting thousands of signals into meaningful investigations is slow, manual, and requires deep domain expertise.

What if AI agents could orchestrate those investigations instead of simply answering questions? In this session, I'll share how we built an enterprise Security Graph Copilot for Microsoft's CISO organization using the Microsoft Agent Framework. Rather than relying on a single LLM, the solution coordinates specialized agents that understand user intent, plan investigations, retrieve and reason over security graph data, invoke domain-specific tools, and synthesize actionable results.

You'll learn the architecture behind a production multi-agent system, the design patterns that make agent orchestration reliable, approaches for integrating enterprise data sources and tools, techniques for grounding LLM responses with graph-based context, and the engineering tradeoffs we encountered while building an AI assistant for real-world security operations. Whether you're building enterprise copilots or exploring agentic AI, you'll leave with practical patterns you can apply to your own AI applications.