Nvidia’s Open Model Super Panel Made a Strong Case for Open Agents
Nvidia's packed panel shifted focus from open models to the systems that orchestrate them.
Nvidia's GTC 2026 'Open Model Super Panel,' moderated by CEO Jensen Huang, became a platform to declare a major industry shift. While titled about open models, the core argument centered on moving 'from models to systems.' Huang dismissed the simplistic 'proprietary vs. open' debate, stating AI is a stack where both coexist. The consensus was that while open models are crucial infrastructure, the greater value and center of gravity is moving to the system layer: orchestration, memory, tools, and runtime that turn raw models into functional intelligence.
Panelists provided concrete frameworks for this shift. Perplexity CEO Aravind Srinivas described 'Perplexity Computer' as an abstraction where the system, not the user, decides which models and tools to use for a task. LangChain CEO Harrison Chase introduced the concept of 'harness engineering'—everything around the model, like sub-agents and memory, that transforms general capability into useful applications. This challenges the notion that such systems are 'just wrappers,' positioning them instead as the future operating system for AI. The discussion, with OpenClaw as a backdrop, signals that routing and agent orchestration are becoming durable, critical product layers.
- Jensen Huang framed AI as a stack where 'proprietary and open' models coexist, moving beyond a binary debate.
- Aravind Srinivas positioned 'Perplexity Computer' as a delegation system that autonomously routes tasks to the best models and tools.
- Harrison Chase defined 'harness engineering' as the critical layer of orchestration and memory that makes AI models practically useful.
Why It Matters
For builders, it means competitive advantage will come from agent systems and orchestration, not just access to the best base model.