Why LLMs Alone Can't Run Your AI Infrastructure | Randolph Chung, MantisGrid AI

LLMs are good at many things, but they can't run your infrastructure on their own. You wouldn't let an LLM drive a self-driving car. In the same way, reliable AI infrastructure needs models that understand how your systems actually work. In this keynote at the MantisGrid AI Hackathon & Summit 2026, Randolph Chung, Co-Founder & CTO of MantisGrid AI, explains how MantisGrid AI turns full-stack signals into decisions and actions. Then he shows it live. A GPU in an LLM inference cluster starts to fail, and within seconds MantisGrid AI finds the root cause and moves the workload to a healthy GPU, with no human action. Recorded September 17, 2026, at Mitchell Park Community Center in Palo Alto, CA.