The room for NVIDIA's Open Model Super Panel at San Jose Civic was packed before Jensen Huang really got going. It felt less like a conference panel and more like one of those sessions where the industry starts saying the next big thing out loud.
NVIDIA billed it as "Open Models: Where We Are and Where We're Headed," moderated by Huang on Wednesday, March 18 at GTC 2026.
But the most interesting argument onstage had almost nothing to do with open models.
It was about open agents (AI systems that can actually take actions, use tools, and complete tasks on your behalf, not just generate text).
- First up, the TL;DR
- The real story: from models to systems
- Srinivas gave the clearest product vision
- Chase made the case for the "harness"
- OpenClaw mattered more as a signal than a product
- Truell connected coding agents to the rest of the economy
- Murati made the strongest case for why openness matters
- Huang's framing tied it all together
- The real takeaway
First up, the TL;DR
The room at GTC 2026 was packed before Jensen Huang started talking, and the panel delivered. Five of the biggest names in AI, including Perplexity's Aravind Srinivas, LangChain's Harrison Chase, Thinking Machines' Mira Murati, and Cursor's Michael Truell, sat down to discuss open models. They ended up making the case for something bigger: open agents.
Huang set the tone early. AI isn't one model, one product, or one winner-take-all category. It's a stack, a system, and increasingly a combination of many models working together.
Here's what stood out:
- Srinivas described "Perplexity Computer" as a system that takes your task and figures out which models, tools, and workflows to use. You stop picking models. The computer picks for you.
- Chase coined "harness engineering" to describe everything wrapped around a model: which sub-agents are used, how memory works, what tools are attached. His point: when you're impressed by an AI product, you're usually responding to the harness, not just the model.
- Murati argued open models are scientific infrastructure, not second-tier alternatives. They widen who gets to build, experiment, and discover new applications.
- Truell connected coding agents to every other industry. Coding was the first domain where agentic AI worked visibly. The same pattern is spreading to research, healthcare, legal, and operations.
The big takeaway: the conversation is shifting from "which model is best" to "who builds the best system around many models." Open models are becoming raw material for specialized intelligence. Open agents are becoming the interface through which that intelligence actually acts. And the harness (the orchestration, memory, tools, and governance wrapped around both) is where the real product value gets built.
The best model might not win. The best system probably will.
The real story: from models to systems
Huang opened the panel by trying to kill the most boring framing in AI: the idea that the market is a clean fight between proprietary labs on one side and open challengers on the other.
His point was broader. AI is a stack, a system, and increasingly a combination of many different model types working together. Not one model. Not one product. Not one winner-take-all category.
"Proprietary versus open is not a thing. It's proprietary AND open," Huang said.
That was the throughline.
Yes, the panel covered open models as infrastructure. Yes, it touched on why open systems widen access and why smaller teams may create some of the most important specialized breakthroughs. But the stronger consensus was that the center of gravity is moving up the stack (further away from raw model intelligence and closer to the software that orchestrates it).
Models matter. Open models matter a lot. But what increasingly matters more is the system wrapped around them: orchestration (how different AI models get coordinated), memory, tools, identity, governance, and runtime.
That's why the panel landed as such a strong case for open agents.
Srinivas gave the clearest product vision
The sharpest product framing came from Perplexity CEO Aravind Srinivas. He described Perplexity Computer in a way that captures where the whole market seems to be heading.
Instead of asking users to choose a model, route tasks manually, and stitch together their own workflows, the system should take the task and figure out how to solve it.
"A.I. is not the model, it's the system. It's the computer," Srinivas said.
That's a bigger idea than product branding.
It suggests the next useful layer in AI might not be a chatbot or even a single frontier model. It might be a computer for delegation: a system that knows which models to call, which tools to use, when open models are good enough, when closed models are worth the cost, and how to pull all of those pieces into one coherent workflow.
Think of it like this: you don't pick which chip processes each tab in your browser. The operating system handles that. Srinivas is describing that same abstraction for AI.
Srinivas also made clear that the future probably isn't a simple ideological split between open and closed systems. Different models will serve different functions. The system decides which one to use, not you.
Chase made the case for the "harness"
If Srinivas provided the clearest product vision, LangChain CEO Harrison Chase provided the clearest builder vision.
His phrase, "harness engineering," might have been the most important term on the panel. Chase used it to describe everything around the model: which sub-agents are used, which skills are attached, how memory works, what tools are selected, and how the environment is configured for a specific domain or task.
"Harness engineering is everything around the model," Chase said.
His point: when people are impressed by a polished AI product, they're usually responding to the system surrounding it, not just the raw model quality.
That matters because it cuts against one of the laziest ideas in AI discourse: that anything built around a model is "just a wrapper." Once models get good enough, the wrapper stops being a wrapper and starts becoming the operating system.
The harness is where general intelligence becomes useful intelligence.
That also helps explain why routing and orchestration (the layer that decides which AI model handles which part of a task) are starting to look like durable product layers. A useful reference point: The Neuron's writeup of OpenRouter. While not identical to what the panel discussed, it maps closely to the same underlying shift. Value is moving into the layer that decides how intelligence gets assembled and deployed.
OpenClaw mattered more as a signal than a product
OpenClaw hovered over the whole conversation, even when the panel wasn't explicitly talking about it.
For context: OpenClaw is NVIDIA's open-source agent framework. Think of it as an open recipe for building AI systems that can take actions, not just answer questions.
Huang framed it as a turning point. In the panel, he called it "a big deal." In a separate GTC press Q&A, he went even further, calling it an inflection point for what comes after reasoning systems. He argued it now needs enterprise-grade layers around it: privacy, governance, security, and optimized runtimes.
The point isn't that OpenClaw is the only product that matters.
The point is that it signals the conversation has shifted from answering to acting. AI systems are moving beyond responses and into execution across files, tools, workflows, and goals. That's the bigger category change the panelists kept circling, even when they used slightly different language to describe it.
Truell connected coding agents to the rest of the economy
Cursor CEO and Founder Michael Truell offered one of the cleanest bridges from coding agents to every other industry.
His argument: coding was simply the first place this style of system started working in a real, visible way. The same pattern is now spreading into other domains.
"What started working in coding last year … now, we're going to all of these other domains," Truell said.
That's a useful lens for understanding why this panel mattered.
Coding agents are the preview, not the final destination. The combination of models, files, CLIs (command-line tools that let you interact with software through text commands), tool use, and rapid iteration made coding the first environment where agentic systems felt obviously real.
If those same building blocks spread into research, healthcare, legal workflows, operations, and back-office work, then the real market isn't "AI coding." It's the much larger category of computer work being reinterpreted as agent work.
Murati made the strongest case for why openness matters
Thinking Machines Founder and CEO Mira Murati gave the deepest argument for open systems. She pushed back on the idea that open models are somehow inherently second-tier, arguing that the gap people see today might be temporary rather than permanent.
More importantly, she framed openness as a way to widen access to research, experimentation, and meaningful technical contribution beyond the biggest labs.
"There is nothing fundamentally different between an open and a closed model," she said.
That moves the discussion beyond the usual "open is cheaper" talking point.
Murati's argument positions open models as scientific infrastructure: a way to expand the number of people who can build, test, specialize, and discover new applications. Even if the biggest labs have a lead today, open systems widen the innovation surface area. More people building means more unexpected breakthroughs.
Huang's framing tied it all together
Huang's role on the panel was more than moderator. He supplied the framing device that made all the other comments click.
He argued that open models, taken together, are already enormous in aggregate and will become even more important as AI spreads across more domains, products, and industries.
That's a useful way to think about the market. The AI market is expanding into too many niches, workflows, and sectors for one type of model strategy to dominate everything. The future is too broad and too specialized to be served by a tiny number of giant systems alone.
The real takeaway
Open models were the headline. Open agents were the case being made.
Open models are becoming the raw material for specialized intelligence. Open agents are becoming the interface through which that intelligence acts. And the harness around both is becoming the layer where trust, customization, product value, and defensibility get built.
The debate is shifting from models to systems.
That's what made this panel feel bigger than a panel. It felt like the industry admitting, in public, that the next AI platform might not belong to whichever lab builds the best single model.
It might belong to whoever builds the best open agent system on top of many of them.
And honestly? That's way more interesting than another benchmark war.
