Speaker: Junde Wu

Institution: University of Oxford

Date: 14:00 GMT, 2nd April, 2026

Talk Title: Human, Agent, and One Context

Abstract: In 2010, software was eating the world. In 2026, agents are eating software. But what is an agent? John Locke argued that identity is continuity of memory. Remove memory, and the self dissolves. Today’s agents are born at the first prompt and die at the end of a session. We are not building beings, we are summoning ghosts. This talk proposes a different view: Agent is Context. Not models, but the continuous thread of memory that gives coherence over time. The core problem of the agent era is not intelligence, it is identity. If memory becomes persistent, shared, and networked, agents stop being tools and begin to form something enduring. In the end, the deepest question is not how smart the model is. It is: who, or what, persists?

Bio: Junde Wu is a PhD researcher in AI at the University of Oxford. His research spans machine learning, agent architectures, and medical AI, with over 50 publications at top venues including CVPR, ICLR, ACL and TPAMI, accumulating 3800+ citations. He was nominated for the CVPR Best Paper Award and won AAAI Most Influential Paper Award. His current work centers on OneContext, an Agentic Operating System grounded in a idea: an agent is not defined by its model, but by its context.