Dr. Ali Arsanjani is the Director of Applied AI Engineering at Google Cloud, and Head of GenAI Blackbelts leading the AI Center of Excellence, within the Google Cloud AI Product Engineering organization.  GenAI Blackbelts are a combination of forward-deployed engineers, researchers and solution architects working with strategic Google AI customers/partners to bridge the delta between emerging customer needs and current product capabilities.  Ali leads strategic co-engineering programs that focus on adoption of Google AI for customers & partners through realizing cutting edge thought leadership, engagement, enablement and execution. Google’s earliest adopters of GenAI have partnered closely with Dr.Arsanjani and his team on their use cases and adoption. He leads Generative and Agentic AI initiatives that enable customers and partners with the best-practices and tools that bring the capabilities of the Google Cloud AI platform to solve tactical and strategic challenges through best-practices and assets and achieving strong partnerships through strategic product co-innovation. Dr.Arsanjani is Senior Lecturer at San Jose State University and the University of California, San Diego, where he teaches and advises students in the Masters in Data Science program and the Data Science Institute.

Engineering Trust: Harness Architectures for Governing Autonomous AI Systems and Solving the Autonomy-Governance Paradox The evolution of artificial intelligence from generative models to autonomous agents—systems that act, decide, and modify real-world environments—necessitates a fundamental shift in control paradigms. Traditional methods like prompt and context engineering are insufficient for managing systems with escalating autonomy. This presentation introduces Harness Engineering as the critical discipline, focusing on the architectural scaffolding (context delivery, tool interfaces, sandboxes, and verification loops) required to govern these high-agency systems. We will explore the Autonomy-Governance Paradox: the more autonomy an agent requires to be useful, the more stringent and automated its governance must become to ensure safety and policy adherence. Resolving this paradox requires moving from model-focused governance to behavior-focused, built-in enforcement. The solution is realized through a Two-Plane Architecture that separates the agent's operational workspace (Data Plane) from the policy enforcement layer (Control Plane).