Back Research Notes OpenClaw and the Arrival of the Agentic Economy: Why Karpathy’s Speed Signal and Huang’s Future Vision Matter Now Published on March 23, 2026 ∙ Download the PDF Report By Jordi Visser We are crossing a threshold that investors are still dramatically underestimating: the end of the chatbot era, where AI merely generated answers, and the beginning of the agentic economy, where AI autonomously executes work. The artificial intelligence community has spent the past two years debating when the “agentic” era would truly arrive. There has been no shortage of hype, prototypes, or bold forecasts. What has been missing is a signal strong enough to separate concept from reality. Last week, two people put into context where we stand right now. In two separate public appearances, two of the most respected people in artificial intelligence, Andrej Karpathy and Jensen Huang, each built a meaningful part of their discussions around the importance of OpenClaw. That matters. It matters not simply because OpenClaw is a new open-source framework for agent orchestration, but because of who is emphasizing it, how early they are doing so, and what it implies about the state of AI adoption. When both a leading practitioner like Karpathy and the most important infrastructure architect in AI, Jensen Huang, center their comments around the same new orchestration layer in separate events, investors need to focus on it as part of their portfolios. It is a strong sign that the agentic world is no longer an idea being marketed forward. It is becoming a real operating model now. That is the real takeaway. OpenClaw is not important only as a product or project. It is important as a signal. It suggests the AI industry has moved beyond the phase where models simply answer prompts and into a phase where systems can manage workflows, coordinate tools, loop through reasoning steps, and produce work more autonomously. In other words, AI is shifting from being a conversational software layer to becoming an operational layer. That is a major change in both technology and economics. AI progress and capabilities continues to blow past adoption. For investors, economists, company leaders, politicians, voters and parents, the world of the past has never been less useful in forecasting the future we are head into. For investors, this is where the story becomes actionable. Karpathy helps explain the speed of the shift. Huang helps explain the scale of the future it creates. OpenClaw sits in the middle, acting as one of the clearest pieces of evidence that the transition from chatbot AI to agentic AI is happening faster than many expected. Karpathy is especially important here because he entered this period from a position of skepticism. In late 2025, he was publicly cautioning against the idea that AI agents were already ready for prime time. His argument was not anti-AI. It was grounded in engineering reality. The systems were still too unreliable, too weak at memory, too inconsistent in action, and too dependent on human supervision to deserve the sweeping claims being made about them. He was effectively saying that the industry was trying to leap from impressive demonstrations to full-blown deployment without respecting the gap in between. That is why his more recent framing carries so much weight. Karpathy’s recent discussion with Sarah Guo on the No Priors podcast reflected a world that had changed quickly enough to alter even his own workflow expectations. And early in that conversation, he highlighted OpenClaw as a driver of that shift. That should not be treated as a passing reference. It should be understood as evidence that the orchestration layer, the part that allows multiple tools, reasoning steps, and agents to work together in loops, has become important enough to sit at the center of how top practitioners think about the future of work. Karpathy is not known for casually endorsing hype. If he is emphasizing OpenClaw, it suggests the environment has changed enough that the orchestration problem is no longer theoretical. It is now central. Jensen Huang, coming from a very different angle, reinforced the same message. At GTC 2026, Huang laid out a sweeping vision of AI infrastructure, inference economics, and what he sees as the industrialization of intelligence. But crucially, he also highlighted OpenClaw. That matters because Huang’s role in the AI ecosystem is not to comment on surface trends. He is architecting the hardware, systems, and economics underneath the next generation of AI deployment. If Huang is spending time emphasizing OpenClaw, then it means he sees agent orchestration not as a niche software layer, but as a foundational development that will drive infrastructure demand. This is why the convergence is so important. Karpathy and Huang are not merely agreeing that AI is advancing. They are pointing to the same emerging mechanism. Karpathy highlights the practical shift in how intelligent systems are used. Huang highlights the macroeconomic and infrastructure consequences of deploying those systems at scale. OpenClaw is one of the bridges between those two worlds. It is part of what makes the agentic era operationally real. The market should take this seriously because agentic AI changes the architecture of demand. A simple chatbot interaction is one inference event. An agentic workflow is something very different. It involves planning, calling tools, querying databases, checking results, using memory, coordinating steps, and often re-running parts of the process. That means it consumes more tokens, more compute, more networking, more storage activity, more CPU orchestration, and more infrastructure intensity than a single question-and-answer interaction. It is a different economic object. That is where Huang’s GTC message becomes especially powerful. His reframing of data centers as AI factories gives investors a language for understanding what agentic AI means in economic terms. Energy goes in. Tokens come out. Atoms into Bits. Those tokens are no longer just generating content; they are increasingly enabling reasoning, task execution, and autonomous action. In that world, the important metric becomes not just raw compute, but the efficiency with which intelligence can be produced and deployed. This is why Huang emphasized inference and tokens-per-watt. It is also why a framework like OpenClaw matters: it helps convert raw model capability into repeated, real-world economic activity. OpenClaw therefore should not be viewed as just another open-source project. It is better understood as part of the enabling fabric of the agentic economy. Its importance comes from helping AI systems move from isolated model outputs to ongoing, structured workflows. That is what makes it so significant that both Karpathy and Huang chose to spotlight it. The timing tells us something. OpenClaw is still new. For something that new to be referenced so prominently by two of the field’s most respected voices suggests that the market may still be underestimating how quickly the agentic layer is becoming real. I hear it every day in my conversations with investors but also with business leaders. Less than 30% of people I speak with inside the investing world see the importance of OpenClaw. There are still people more people than you would believe saying all AI does is predict the next word. It’s just autocomplete on steroids. This also helps explain why the AI investment opportunity is broadening beyond the narrow GPU story and the reward to risk may seem too obvious. Don’t underestimate the skeptics impact on the odds that exist in the market for the broadening out of AI. If OpenClaw and similar frameworks increase the deployment of agents, then demand broadens across the whole infrastructure stack or the “Whole Rack” as I published last week. Agentic systems require not only accelerated compute but also CPU coordination, memory, data movement, storage retrieval,