Jensen Huang thoroughly explained the Agent architecture—LLMs as the brain, orchestration engines as the OS, tools available on demand. After CUDA X is fully open, efficiency surpasses humans. The software industry isn't the end of days but an upgrade. The part where Claude Code writes code on the spot is indeed smooth.

Nvidia's Jensen Huang discussed the AI Agent architecture at GTC Taipei 2026: making large language models responsible for thinking and planning, with external orchestration engines acting like operating systems, connecting tools such as spreadsheets, browsers, databases, and managing memory, driving a new computing paradigm of "describing intent, generating code, calling tools, producing results." The demonstration included generating application code, dynamic particle animations, CAD printing replacements, etc., emphasizing the strength of Claude Code/Codex; he stated that CUDA X is now fully open to Agents, with efficiency even surpassing human developers, and claimed that massive numbers of Agents will bring better development opportunities to the software industry rather than causing its collapse.