What happens when you turn the tables on your own AI? In this session, we dive into the fascinating journey of building an autonomous agent with Copilot Studio—and then trying to break it. You’ll hear the real story of how an agent designed to streamline job applications was stress-tested with prompt injections, tool misuse, and data exfiltration attempts. Discover where the agent held strong, where it failed spectacularly, and what lessons emerged for anyone building AI-driven automation. This is not theory - it’s a candid look at the vulnerabilities, defenses, and design patterns that matter when your agent moves from the lab to the real world. Why Attend? If you’re working with Copilot Studio, Power Automate, or planning to deploy autonomous agents, this session will give you practical insights into: • Common attack surfaces and how to mitigate them • Designing safe, auditable actions and flows • Grounding AI responses to enterprise data • Building resilience without sacrificing usability Join us for a story that blends innovation, security, and a little bit of hacking fun—because the best way to understand your agent is to try breaking it.

Responding to RFPs is a complex, high pressure process that rarely fits into a single workflow—or a single AI agent. Requirements analysis, compliance checks, content drafting, and consistency reviews all demand different skills and perspectives. In this session, I present Proposal Operator, a practical case study of using multiple Copilot Studio agents working together to support RFP responses. Instead of building one “do everything” agent, the solution is based on specialized agents with clear responsibilities, orchestrated as a team to handle different parts of the process. The focus is not on clever prompts, but on architecture and orchestration: when one agent delegates work to others, how responsibilities are separated, and how the system stays understandable and governable as complexity grows. The RFP scenario serves as an example of a broader pattern relevant to many enterprise processes where work must be split, reviewed, and coordinated. You’ll see why multi agent designs are often more robust than single large agents, what breaks when orchestration is missing, and how Copilot Studio can be used to model agent teamwork in a controlled way. This session is for architects and developers who want to move beyond single agent demos and start designing agent systems that scale in both capability and complexity.