A Deep Dive on Nvidia’s Agentic AIA Deep Dive on Nvidia’s Agentic AI

In his announcement-filled keynote address to kick off this year’s CES consumer electronics event in Las Vegas, Nvidia founder and CEO Jensen Huang introduced a flurry of new and enhanced products. I’ve covered several of them in previous articles. I want to focus on one of his most interesting and impactful announcements: agentic AI. 

The age of agentic AI is here, and every organization can take advantage of it. Huang called it “a perfect example of test-time scaling,” which refers to a computer, or in this case, an AI that can pass the famous Turing Test designed to distinguish human thinking from its machine counterpart. 

What is agentic AI, and what will it enable? 

Let’s first define agentic AI and then examine its potential benefits for the world. Nvidia calls agentic AI “the next wave in the evolution of generative AI.”  

Agentic AI is a system of AI models or agents that can act autonomously, make decisions, and learn independently. Agentic AI works with limited human supervision to perform a wide range of tasks that, until now, could only be done by people. Huang described AI as a system of models. “Maybe it's going onto the internet. Maybe it's starting with a PDF file. And so it might be using tools. It might be using a calculator, and it might be using a generative AI to generate charts and such. And it's taking the problem you gave it, breaking it down, step by step. And it's iterating through all these different models,” he said. 

Related:Cisco Report: Enterprises Ill-prepared to Realize AI’s Potential

Since the dawn of the computer age in the mid-20^th century, humans have driven the processes that computers rely on to perform work and output responses. “In the future, you’ll ask a question and a whole bunch of models are going to be working in the background,” Huang said. “The amount of computation used for inferencing is going to go through the roof because we want better and better answers to help the industry build agentic AI.” 

How will agentic AI reach the market? 

Huang said Nvidia’s go-to-market approach is to “work with software developers in the IT ecosystem to integrate our technology to make possible new capabilities, just like we did with CUDA libraries. We now want to do that with AI libraries.” There will also be what Huang called exciting new models coming from physical AI. “These AI models run in every single cloud because Nvidia's GPUs are now available in every single cloud. It's available in every single OEM,” he said. Huang explained that the models could be integrated into a wide variety of software packages to create AI agents on many platforms for customer deployment. 

Related:Business Continuity Depends on the Intersection of Security and Resilience

Huang said the next layer of agentic AI development will leverage the company’s end-to-end platform for developing custom generative AI, including large language models (LLM), vision language models (VLM), video models, and speech AI. 

“These AI agents are essentially a digital workforce working alongside your employees, doing things for you on your behalf,” he explained. Huang added that companies would onboard these specialized agents similarly to how new employees are onboarded when they join a company. “We have different libraries that help these AI agents be trained for the type of language in your company. Maybe the vocabulary is unique to your company. The business process is different. The way you work is different. So, you would give them examples of what the work product should look like, and they would try to generate it, and you would give it feedback, and then you would evaluate them and so on.” Huang said companies would give the agents access to certain information, but there would be guardrails to prevent agents from doing anything the company doesn’t want them to do. 

Huang said the agents would be a digital employee pipeline, resulting in a company’s IT department becoming “the HR department of AI agents in the future. Today, they manage and maintain a bunch of software from the IT industry. In the future, they will maintain, nurture onboard, and improve a whole bunch of digital agents and provision them to the companies to use.” 

Related:Infrastructure Insight: Resilience is Uptime’s Secret Sauce

Nemotron model families will advance agentic AI 

Nvidia will provide a foundation for agentic AI based on its Llama Nemotron family of open LLMs. These models will help developers build and deploy agents for a wide range of applications, including customer support, fraud detection, product supply chain, and inventory management optimization. 

Huang said Llama 3.1 has already been downloaded from Meta hundreds of thousands of times and turned into about 60,000 other different models fine-tuned for enterprise use. “It is singularly the reason why just about every single enterprise in every single industry has been activated to start working on AI.” 

Next up: AI PCs 

Huang said AI was created in the cloud and for the cloud. It’s already on phones, but he said the goal is to put Nvidia AI inside companies to become an AI assistant. He said the way to do that is to put AI directly into PCs. 

The key to achieving that, he said, is on computers with Windows Subsystem for Linux (WSL) 2. “It’s developed for developers” and “optimized for cloud-native applications and CUDA” right out of the box. Nvidia is working to turn WSL 2 Windows PCs into a “target first-class platform” that Nvidia will support and maintain “for as long as we shall live.” He called it “an incredible thing for engineers and developers everywhere” and an example of the blueprints Nvidia and its partners are developing to make this a reality. 

“All of the PC OEMs we work with -- basically all the world’s leading PC OEMs -- are going to get their PCs ready for this stack,” said Huang. “And so AI PCs are coming to a home near you.”