Inside on-device agentic AI on Arm: Google and the Arm heterogeneous compute foundation

Agentic AI is moving mobile experiences beyond generating responses toward systems that can understand intent, gather context, coordinate actions, and help users get things done. In this video, Oli Gaymond, Product Lead, Android AI at Google, shares his perspective on this shift and the heterogeneous computing foundation behind it. CPUs, GPUs, and NPUs increasingly work together across complex AI workflows, with the CPU providing the programmable compute and orchestration needed to coordinate activity across the system. Discover how Arm Scalable Matrix Extension 2 (SME2) expands on-device AI capabilities for latency-sensitive workloads such as speech, retrieval, and model execution, helping enable the responsive, interactive experiences agentic AI demands. This is part of the new Arm CSS for Mobile 2 platform which brings together CPU, GPU, System IP, software, and developer tools, so performance, efficiency, and data movement is optimized across the complete workload. You’ll also hear how Arm’s software ecosystem connects these underlying compute capabilities with the frameworks, models, and development environments developers already use, helping move increasingly capable on-device AI experiences from evaluation to deployment. Learn how heterogeneous compute is helping shape a new generation of on-device AI that can understand context, reason across tasks, and turn intent into action. Stay connected with Arm: https://www.arm.com/