Talk Title: Guardrails for Agentic AI Enterprise

Speaker: Ankita Upadhyay, Senior Director, AI Enablement, Thomson Reuters

2025 marked a definitive shift from Large Language Models to complex Agentic AI solutions. We have moved from asking 'What can AI answer?' to 'What can AI do?', enabling enterprises to embrace autonomous decision-making. Consequently, the primary challenge has become the governance of Agentic Autonomy—managing systems that perform independent planning, utilize integrated tools, and execute multi-step processes without constant human intervention.

This session traces the evolution of Responsible AI from its foundations in LLMs to the critical necessity of Agentic Governance in 2026. We will also explore the urgent need for non-linear forms of accountability, ensuring ethical guardrails scale alongside AI’s increasing agency.

Talk Title: General Visual Intelligence with Video Models

Speaker: Priyank Jaini, Research Scientist, Google DeepMind 

Just as Large Language Models (LLMs) transformed natural language processing from bespoke tools into unified foundation models, this talk argues that video generation models are poised to drive a similar paradigm shift in computer vision. The talk will demonstrate that large-scale video models, such as Veo 3, possess emergent zero-shot capabilities that extend far beyond video generation. Without any task-specific training, these models can perform tasks across the entire vision stack—from fundamental perception (e.g., segmentation, edge detection) and physical modeling, to image manipulation and complex visual reasoning.