The Next Frontier

The Next Frontier An Evening with the People Shaping What's Next in AI A private House of 434 gathering at Snowflake, Menlo Park We are entering a new chapter of artificial intelligence. In just a few years, AI has moved from generating text and images to writing software, solving increasingly complex problems, conducting research, and carrying out work that once required teams of people. Yet for all the progress, some of the most consequential questions remain unanswered. Are today's architectures sufficient to take us to the next level of intelligence? What comes after the current generation of large language models? How much autonomy can we safely give AI systems? What happens when intelligence becomes abundant, but compute, energy, data, and trust remain scarce? And perhaps most importantly, what becomes possible to build now that wasn't possible before? The Next Frontier brings together researchers, founders, engineering leaders, and investors at the forefront of AI for an evening of candid conversations about where the field is heading. We will explore the breakthroughs happening inside leading AI labs, the systems bringing those capabilities into the world, the technical and economic constraints that could shape the next decade, and the opportunities emerging for founders. This isn't a discussion about the latest product announcements or incremental improvements to existing tools. It's about the deeper shifts in intelligence, computing, and company creation that could define what comes next. Speakers: Sriram NatarajanEngineering Leader, Gemini Enterprise, Google Sriram works on enterprise AI at Google, tackling complex workflows and the challenges of bringing AI into production. Previously, he was Senior Director of Engineering, Applied AI at GEICO and Co-founder & CTO of Yuni. He brings experience building AI systems and addressing the reliability, governance, and deployment challenges that come with greater autonomy. ****Additional speakers to be announced.**** The Conversation 01 / Beyond Today's Frontier Models Are we approaching the limits of current approaches, or just beginning to understand their potential? Scaling has driven extraordinary progress in AI, but the next breakthroughs may require more than larger models and more compute. We will explore the research directions that could reshape the field, from new reasoning and learning approaches to multimodal intelligence, memory, adaptation, and more efficient architectures. What are the most significant limitations of today's frontier models? Where are additional compute and scale still delivering meaningful gains, and where might new approaches be necessary? Can AI systems learn continuously and adapt to unfamiliar environments? What research directions are underappreciated today but could become foundational over the next five years? 02 / From AI Assistants to Autonomous Systems What happens when AI stops waiting for instructions and starts carrying out complex work? AI is moving beyond individual prompts toward systems that plan, use tools, coordinate with other agents, and execute longer sequences of work. But autonomy introduces new challenges in reliability, verification, safety, cost, and control. We'll examine what it takes to move from impressive demonstrations to systems capable of operating independently in consequential environments. How close are we to AI systems that can reliably execute complex, multistep tasks? Will multi-agent architectures become fundamental to how software is built, or introduce more complexity than value? What must change in memory, evaluation, orchestration, and oversight? Where should humans remain in control, and where could autonomy fundamentally change how organizations operate? 03 / Intelligence Beyond the Screen What changes when AI must understand and act in the physical world? Much of the recent AI revolution has happened in digital environments. The next wave of applications may increasingly involve the world beyond them. From robotics and industrial automation to scientific laboratories, semiconductor design, energy systems, and advanced manufacturing, AI is beginning to interact with physical processes and infrastructure. The opportunities are enormous, but so are the challenges. Physical environments introduce uncertainty, limited data, real-time constraints, safety requirements, and consequences that cannot simply be undone. What will it take to move from digital intelligence to systems that understand and operate in the physical world? Where are foundation models genuinely changing robotics, science, and industrial automation? How important are simulation, world models, and real-world data? Which applications are approaching commercial viability, and which still require major research breakthroughs? 04 / The Infrastructure and Economics of Intelligence If intelligence becomes abundant, what remains scarce? The economics of AI are changing rapidly. Training and inference costs, compute availability, energy consumption, model efficiency, and infrastructure constraints are shaping which technologies can scale and which companies can survive. As models become more capable and accessible, the competitive advantage may shift toward entirely different parts of the technology stack. Will compute remain the defining bottleneck, or will energy, memory, data, and system efficiency become equally important? How will improvements in inference economics change what developers can build? What happens to differentiation when powerful models become broadly available? Where will enduring value accumulate: models, infrastructure, proprietary data, applications, distribution, or entirely new layers? 05 / The Next Generation of AI Companies What would you build today if you were starting from zero? Every major technology shift creates a new generation of companies, but the greatest opportunities are rarely obvious at the beginning. As AI capabilities advance, existing products, business mod