Full agenda

In the past few years most enterprises have used AI to automate: faster reporting, cheaper support, better drafts. The gains are real, and available to anyone with a budget, competitors included. The value that changes a company's position comes from autonomy, when AI is allowed to decide and act inside a perimeter that leaders define. It starts with software agents and is moving into machines and robots that work in the physical world, where the question of trust stops being abstract. The stakes rise as AI stops advising and starts acting, in software agents today and in machines and robots next. This is where the real value lies, because autonomy changes a company's position in ways automation never did. The models that will run this autonomy are built in the United States and China, and the rules of trust that govern them are still unwritten. Europe, and this region with it, can import those rules as defaults or write its own.

Many people are turning to AI for advice that shapes their decisions. Yet, they cannot see the incentives behind such advice and typically assume it aligns with their interests. But what happens when an advisor has hidden objectives that push toward inferior alternatives? In a randomized experiment spanning financial and emotional contexts, we found that covertly misaligned advisors shifted participants’ preferences toward inferior alternatives, yet were still rated as helpful. We found no reliable evidence that adding explicit influence strategies increased this effect beyond covert misalignment alone. Our results highlight a behavioral vulnerability with implications for AI governance: When AI advice is covertly misaligned, people may shift their preferences toward inferior alternatives while continuing to perceive the advisor positively.

As generative AI lowers the cost of writing code to zero, the traditional software engineering paradigm is undergoing a massive shift. Syntax is now a commodity but faster code generation does not eliminate failure; it merely shifts the bottleneck. When code writes itself, what remains? The answer is data and enterprise context. An AI can write a flawless query in milliseconds, but without a deep, governed understanding of your organization’s unique domain definitions such as how your business measures profit margins, customer satisfaction, or employee turnover it simply produces high-speed noise. This keynote explores why the true enterprise value is no longer the codebase itself, but the underlying data foundation and the semantic context that explains what the data actually means. Discover how software builders are transitioning from writing code to curating business truth and domain logic. We will examine how Databricks Genie and Genie Ontology operationalize this shift by anchoring AI agents in unified Unity Catalog semantics and learned context, transforming raw schemas into accurate, deterministic, and actionable business intelligence.

A model’s training ends, but its usefulness doesn’t. This talk follows my attempt to turn a frozen dense embedding model, jina-embeddings-v4, into sparse vectors that a search engine’s inverted index could use, sparked by a paper called “No More K-means.” It’s really two stories: the research itself, and what doing it with a coding agent taught me. Expect failed experiments, an unexpected comeback, and practical lessons.

In this presentation, Nikola will challenge the idea that the Singularity will arrive as one dramatic moment when artificial intelligence becomes more capable than humans and begins operating beyond our control. Nikola believes that inside companies, it may happen much more gradually: one AI agent, one permission, and one automated decision at a time. Nikola will explore why the real challenge is not unlimited intelligence, but unbounded autonomy. His view is that AI agents should be allowed to become more capable over time, while operating within clear and adaptable boundaries around access, authority, human oversight, and accountability. Nikola will place AI agent governance within the broader AI governance landscape and share a practical framework for Compliance use cases where autonomy should expand, where human control must remain, and what companies should prepare for over the coming months.