Jensen Huang defines a new AI computing paradigm: Agent = Large Model + Orchestration Engine, ushering in a golden age for software companies.

According to Beating's monitoring, Jensen Huang systematically explained the architecture of the "AI Agent" in his GTC Taipei 2026 presentation: a large language model is responsible for thinking, reasoning, and planning, while an external orchestration engine (harness) acts like an operating system, connecting the model with tools such as spreadsheets, browsers, and databases, and managing working memory and long-term memory. He defined it as a completely new computing model—in the past it was "launching an application and clicking to input," in the future it will be "describing your intention to AI, which generates code, calls tools, and produces results." Huang demonstrated several Agent examples on-site: generating complete application code using natural language prompts; generating dynamic particle animations of Taipei 101 → GTC Taipei 2026 → NVIDIA logo from a given description; and even taking a picture of a lost battery compartment on a remote control, with the Agent directly calling CAD tools to generate 3D-printable replacement part files. He mentioned, "We're using Claude Code here, but Codex performs equally well." In response to the claim that "Agents will bankrupt software companies," Jensen Huang refuted it again: "Quite the opposite." He believes that when the world is no longer limited by the human population, massive numbers of agents will use more software tools than humans, and "now is the best time to start a software company"—but only if the software is presented in a way that agents can invoke. He also announced that NVIDIA's CUDA X libraries are now fully open to agents, and agents are even more efficient at using these libraries than human developers.