Agentic AI for Every Industry | GTC Live Washington, D.C. Chapter 2

The podcast explores the transformative potential of agentic AI across various industries, emphasizing the shift from research to real-world applications. Panelists discuss the importance of scaling American technology globally, with the U.S. leading in AI advancements. They highlight the emergence of AI agents that enhance productivity and user engagement in both business and daily life.

Eravon Srinivas from Perplexity envisions their browser, Comet, as a personal assistant that boosts user interaction by allowing questions from any platform. Scott Wu of Cognition notes the rapid growth of coding agents in enterprises, addressing the gap between AI hype and actual productivity gains, particularly in software engineering.

Shiv Rao highlights challenges in healthcare, where many clinicians are considering leaving due to operational inefficiencies. He advocates for AI to alleviate clerical burdens, allowing healthcare professionals to focus on patient care. George Kurtz discusses the evolving landscape of cybersecurity, emphasizing the need for AI-powered automated security operations to manage increasing threats.

The panel expresses excitement about the future of consumer experiences, particularly the transition to more intuitive, answer-based systems. They introduce the concept of a browser agent that understands user preferences and facilitates transactions, marking a significant advancement in user-centric technology.

Concerns about the speed of browser-based agents are addressed, with a focus on asynchronous operation rather than chat formats. The potential for agents to handle multiple tasks simultaneously is illustrated. Security concerns in AI are raised, questioning whether larger companies have an inherent advantage due to data scale, while emphasizing the importance of how data is collected and utilized.

The conversation shifts to Washington's role in enhancing national security in the AI era, focusing on improving procurement processes to avoid outdated technology acquisitions. The current administration is viewed as forward-thinking, aiming to implement advanced technologies and automation for future-proof security measures.

Discussions on AI inference costs reveal that while inference is a significant expense, costs are expected to decrease with better models. Companies are collaborating with Nvidia to create optimized inference libraries, and new subscription models are emerging to support background tasks. The need for advanced reasoning models is acknowledged, despite their high operational costs.

Scott discusses the substantial computational resources required for agents, noting the complexity of tasks that may involve numerous queries. Despite high compute demands, the productivity gains justify the use of GPUs. Models are evolving to be smaller and faster, though there is a trade-off between intelligence and speed.

The conversation centers on the potential of agentic AI systems to replace traditional operating systems like Windows and iOS. Aravind highlights the browser's critical role in deploying personal agents on mobile devices, noting that while Android appears open, it is ultimately controlled by Google. He stresses the importance of addressing software challenges over developing new hardware, expressing optimism about future innovations that could improve user interaction.

The discussion also touches on significant developments from OpenAI and Microsoft in the healthcare sector, prompting the guests to stay informed about these advancements.