NVIDIA CEO Jensen Huang introduced a groundbreaking AI computing paradigm during his GTC Taipei 2026 keynote, defining "AI Agents" as a combination of large language models and an orchestration engine. These agents are designed to handle complex tasks by connecting models to various tools like spreadsheets and databases, managing both short-term and long-term memory. Huang emphasized a shift from traditional application use to AI-driven intent description, where AI generates code, invokes tools, and delivers results. Huang showcased live demonstrations, including generating application code from natural language prompts and creating a 3D-printable part from a photo. He dismissed concerns that AI Agents might threaten software companies, arguing instead that they will expand software usage beyond human limitations. Huang also announced that NVIDIA's CUDA X libraries are now fully accessible to AI Agents, enhancing their efficiency in utilizing these resources.