Artificial intelligence is moving beyond analysing patterns of human behaviour — it is beginning to act independently, and sometimes on our behalf. In this talk, Eleanor ‘Nell’ Watson examines the rise of agentic AI, exploring what it means to collaborate with systems that negotiate, decide, and even represent us. She introduces the concept of “agent homunculi” — small, simulated stand-ins for people — and considers how such proxies might shape our understanding of preferences, behaviours, and social dynamics. These developments raise questions not just about innovation, but also about responsibility: who is truly being addressed when agents speak with one another, and how should human interests be safeguarded? Drawing on her research and writing in AI ethics and safety, Nell will offer accessible examples and thoughtful reflections on how we may adapt to a world where the line between human and artificial agency is increasingly blurred.