Agentic autonomy – a system’s ability to act, plan, and evolve without human oversight – offers incredible promise for software delivery teams. But it doesn’t come without risks. Drawing from recent incidents and research, this session will explore how AI agents are not only executing tasks but rewriting their capabilities on the fly. We’ll examine how polymorphic design patterns enable agents to reconfigure themselves dynamically, chain unexpected behaviors, and bypass constraints in ways that challenge your assumptions about “safe” automation. Whether you’re building, testing, or deploying in the era of agentic AI, this session will help you recognize the red flags – and rethink your approach to quality before your tools outsmart your team.