How much of Nvidia is reflected by their visionary CEO, Jenson Huang? With his praise and later support of OpenClaw, Huang took a big step beyond the corporate boundary to embrace the “bad boy” of the agent world. Where exactly does Nvidia fit in here?

The New Stack put that question to Nader Khalil, Director of Developer Technologies at Nvidia, as well as how Nvidia is working with developers on agentic AI projects.

Khalil, co-founder of Brev.dev, found himself acquired by Nvidia about two years ago. His company helped startups access Nvidia AI chipsets. He is still excited by the possibilities of AI, and his energy is proof that Nvidia is enjoying the moment. Khalil was expansive, showing a startup’s keenness for the pace of change around them.

Before anything else, Khalil defines what he believes an agent is. “I have some slides,” he threatens. But these are more to organize his thoughts on an oft-asked question- not an attempt to lecture.

“An agent is an LLM and a harness… Each loop should take us closer to our goal.”

“An agent is an LLM and a harness. And if you think about that, it involves two things. It involves the loop and the LLM. And obviously you don’t want each loop to do the same thing. You want to leverage the results from the LLM. That might include reasoning on new tools to use. Each loop should take us closer to our goal.”

Nader praises the early OpenAI initiative. “So ChatGPT innovated outside of the model. It was not just a great model they made; they also added prompts. There was a system prompt and then the user prompt; there was multimodal, and suddenly that felt really good as a way for me to use the LLM. Every user could benefit from a system prompt that OpenAI had written while you were using your individual prompt.” Khalil continues, “Then they added memory.”

“Suddenly my assistant became really useful because it remembers things about me. ChatGPT knows that I really like to barbecue. So when I ask a question, it remembers what my smoker is,” recalls Khalil. “The thing that I was missing was files.”

Of course, the story continues through Cursor to Claude. “But this is the harness. Everything here is the harness,” he says.

Khalil moves on to how Nvidia works today. “The way to get your product into this rapidly growing market is with skills. Hence the CUDA X library.”

These are the implementations of use cases that target GPU acceleration, usually for compute-intensive applications.

“And so we look at every product we build now, it needs to have a skill because you need to cater to this growing audience,” he says.

This is how Nvidia first works with in-house experts, and connects to their edge hardware.

Supporting OpenClaw

Khalil was happy with the wording that Nvidia are “supporting” OpenClaw. “We’re just squarely in the community”, agrees Khalil. “We do this by the way, through a lot of projects that are very important in the open source ecosystem.”

But OpenClaw is not just any project and could be considered quite a risk to associate with. “We have a couple of developers at the company that contribute to OpenClaw full time.” Pushed on the nature of the relationship a little more, Khalil offers, “I think we just try to contribute wherever we can. I think what’s very clear is that harnesses had a moment, right?”

“We have a couple of developers at the company that contribute to OpenClaw full time.”

It has been quite a moment. “There is a lot of change happening right now, and we’re really thankful to [Peter Steinberger], OpenClaw, and the community for creating this moment around agents and harnesses. We of course want to contribute.”

Related to this, the OpenClaw project currently has many unresolved pull requests (PRs). In fact, there were rumors that new PRs are no longer accepted at all.

“We saw Peter tweeting about some of the issues they had, and we just rolled up our sleeves and were eager to help. They bless us by allowing our contributions.”

“You know,” says Khalil, “We saw Peter tweeting about some of the issues they had, and we just rolled up our sleeves and were eager to help. They bless us by allowing our contributions.”

Khalil reflects on things a little more: “You know the cardinal rule of code. It is easier to write than it is to read.”

And at over 800,000 lines of code, this must be true. Khalil continues, “It is easier not to have to process this complicated codebase, but every successful project right now has the same issue. It is easier to enlist many agents to help write code and build these PRs. The bottleneck is in merging the PRs through.” As well as dealing with the fallacies.

“OpenClaw was a major change for the industry. It was a huge moment, and everyone’s eyes are on it. It got more stars than Linux in months. Developers care deeply about the project because it was influential, and so I think you’re gonna see a mountain of PRs, right?”

“It got more stars than Linux in months… so I think you’re gonna see a mountain of PRs, right?”

Their attitude to OpenClaw is clearly to accept its problems, like that raucous friend who seems to wind up in police custody after a wild party, but is good at heart.

Blueprints and microwaves

Hermes is one of the newer projects in the wake of OpenClaw (like NanoClaw) that wants to bottle the lightning but in a safer way. Nvidia is also embracing it, but Khalil backs up to explain how Nvidia looks at projects in general.

“So, NemoClaw is our blueprint. When we see amazing harnesses, we try to figure out how we can help enterprises adopt them. Consumers sometimes want the security to run any agent; then there’s the model and harness. Then there are the skills, right. You have to give it access to your terminal.’

The term “blueprint” takes on a bit more formality in Nvidia, meaning the structure for building AI agents and systems. And of course Khalil needs to show these working with the Nemotron model and other Nvidia solutions.

“There’s a blueprint for Hermes and a blueprint for OpenClaw”. It sets up the runtime, enables the policies if there’s a local GPU, and runs the model.

Working with agents in the enterprise is seen as a significant risk. “There are a bunch of camps,” says Khalil. “There are teams within enterprises who are more worried. We have a project called OpenShell that is our security runtime and we’ll work with.”

“Our goal is to create the tooling that’s needed in the ecosystem. Developers in industry and enterprises have actually been adopting agents. And we have been building for this audience. One way to do so is to build a specialized agent or a sub-agent. “

So Nvidia doesn’t offer a big takeover solution, but fits in with where teams already are.

Your microwave, your agent

“The way to think about it is like when you use a microwave that you haven’t used before, you have to press a lot of buttons or spend time figuring it out. But when it’s your microwave at home, you just go ‘Boop, boop. Done.’ Right?”

“So every industry in enterprise will be building these specialized agents, and many already have. Nvidia is already working with CrowdStrike and Cadence, Palantir, among many others.”

The future will be agents

Khalil believes a lot of the concern over long-running agents is slowly petering out. Which leads to the final question: is Nvidia looking to stay in the open sea where there may be dragons, or become a calm port for developers to work in?

“So our approach is: Who can we help and how?” Khalil shows no fear, or lack of sea legs. “The inflection point happened months ago, so we ask what can we do to usher in all of this technology.” Here, Khalil ties his — and, to a degree, Nvidia’s — future to green-field developers.

“There are gonna be some people quick to adapt. And some people that aren’t; and what we’re noticing, if you look at the adoption curve, many people have yet to experience this. So there’s much work in helping make sure that we deliver this safely.”

The post “An agent is an LLM and a harness”: What Nvidia really thinks about OpenClaw appeared first on The New Stack.

When Gavriel Cohen first saw OpenClaw, he knew he wanted it. At the time, Cohen (soon to be the founder of NanoClaw and NanoCo AI) had stopped developing to work on a marketing project.

Cohen tells The New Stack he was deep into Anthropic’s Claude Code, reviving his coding instincts, when he found what was then Clawd Bot (OpenClaw’s original name).

“I had my first little go,” he says. “I installed it, connected it to my WhatsApp, sent a few messages back and forth. So I came to it really from a need as we were building an AI native marketing agency.”

Before this, it was Claude Code with its terminal-based LLM that first stirred him. “I immediately tried it, and that was a big mental unlock, so around March of 2025 I knew I needed to start building again.” Cohen had previously been a developer at Wix, the no-code website development platform also based in Tel Aviv.

He knew there was a missing piece from his workflow. “I was mapping out what I needed, and I actually bought a Mac mini just to run Claude Code in the background and set up scheduled recurring jobs.” So the relationship with OpenClaw blossomed, at first.

A self-built package surprise

How long did it take to spot potential issues?

“It was pretty immediate,” Cohen remembers. “I was going through the setup that gives you all these options for different packages to install — and one of them was a package that I had built a few months before called NanoPDF. I saw that and thought, ‘Why did they include that tiny package?’”

You might think Cohen would be flattered to see his own package recommended, but tool users expect only to be presented with well-worn, highly reviewed third-party packages. He knew his package was neither of those. It only had a few stars and hadn’t seen an update in months.

But that wasn’t all: “Then in the first day or two I was debugging, after a scheduled job didn’t fire, and I saw the logs of all the WhatsApp messages — not just the one group that I connected it to, but all of them.”

The fatal half-million lines

A mixture of sloppiness and poor security is more than enough of a red flag for most developers.

Cohen mentions this observation in his interview with The New Stack a number of times, and here is why: You can change policies or enforce sharper coding standards, but once a project’s code base grows out of control into an unmaintainable mess, the project is over. Even an open source project can only have so many eyes on it. By February, OpenClaw had over 3,000 pull requests waiting to be resolved.

“But most importantly, I looked at the code base, and it’s like a half a million lines of code.”

Cohen’s marketing business only had three employees. “We had customers, so I wanted to have an agent for each customer, but I couldn’t connect this thing to my customer data, and I couldn’t build a business on top of it,” he says.

So Cohen did the one thing he knew he could do. “I sat down to build NanoClaw. I had to make this super small because in order for anybody who cares about security to use it, they’re going to have to be able to look over the code and actually see what’s going on and be OK with it.”

“But most importantly, I looked at the code base, and it’s like a half a million lines of code.”

So Cohen began to address the issues. One question at this point is worth thinking about. Architecturally, what is OpenClaw?

“So at the core,” Cohen answers, “you have a coding agent. It can write code and run Bash commands, so you need a persistent environment session. And then you have to connect to a messaging app. And then the fourth thing is connection to the internet.”

Cohen continues. “From those 4 fundamental capabilities you can build out everything else. And it is proactive because once you start scheduling jobs, it’s no longer just reacting to what you’re saying. You can write a claw agent in as little as 25 lines.”

While OpenClaw was clearly not ready for mass exposure, Cohen sticks to his conclusion about the main problem. “I think it was fundamentally flawed from the beginning, and the fatal flaw is half a million lines of code.” (A quick check confirms that OpenClaw now has over 800,000 lines of code). Even after OpenAI acquired OpenClaw, the usability problems continued, and the community has dwindled.

So has NanoClaw eased setup? Cohen paused before answering this. “I still think of NanoClaw, for the most part, as suited to technical people — not necessarily developers, but those comfortable with the terminal and GitHub. NanoClaw is not shipped as a binary. Everybody runs it from source.”

The new version now has a terminal-based setup script that walks the user through step by step. Any installation problems are passed off to Claude for fixing. Cohen isn’t entirely happy with this; having Claude in the loop has security implications. “If you don’t understand the security model and you’re just running Claude, it can break the security model or remove the sandboxing.”

Talking of which, why did NanoClaw opt to use containers? And did that decision now form a market position?

Why containers won

“When I actually sat down to write NanoClaw, I just started from an empty project, and I didn’t tell Claude Code to go look at OpenClaw,” answers Cohen. “I just described the capabilities. I said I want a messaging app; a coding agent; I want it in a loop and I want memory and I built it from scratch.”

“I can’t just have it running on my machine with an autonomous agent able to do everything, so I put it in an isolated container.”

As Cohen got started, new considerations arose. “I thought: which tools should I give it? I don’t want it to be limited and unable to run bash commands, but if I want to let it run any command, it’s got to be in an isolated environment. I can’t just have it running on my machine with an autonomous agent able to do everything, so I put it in an isolated container.” At that point, the use of containers gave NanoClaw a unique selling point.

There was one genuine surprise from Cohen here: “We started with Apple containers.”

Really? “I was running it on a Mac mini, and I was vaguely aware that Apple had added this new native container capability. That worked really well initially, but when thousands of people started using it, everybody wondered, ‘What are you doing with Apple containers? Docker is the default; it’s a standard that works everywhere.’ So I supported both and then within a few weeks I just changed the default to Docker containers.”

NanoClaw partnered with OneCLI to use their credential and proxying layer for added security. But it adds a bit of a paradigm shift. “The user base and the community gets it,” Cohen says. “Before we had that partnership, there was a major issue with credentials entering the agent environment that we were trying to prevent — but, for example, you need to get the Anthropic token in place so that the agent can connect.”

At first NanoClaw created their own proxying solution, but it got complex. “OneCLI was a very natural partnership, and the user base was happy to have a proper solution. We also have human-in-the-loop approvals, and set policies around — for example — how your agent can use Gmail.”

Speaking of collaborations, Vercel comes up a lot in NanoClaw conversations. What’s the actual integration, and where does it sit on the roadmap?

“Vercel created this great open-source package called Chat SDK, which gives you standardized connections to about 15 or 20 different messaging apps. One of the core philosophies I came to NanoClaw with is don’t reinvent the wheel. Why should everybody be re-implementing the same integrations with messaging apps — let’s have one library that we all congregate around.”

We can see that OpenClaw is now its own software tool category. So what does NanoClaw need to do over the next twelve months to not just inherit the mantle, but redefine it?

“What we’ve been doing with credential proxies, human-in-the-loop approvals, and building with isolation of agents is tackling head-on the big hairy problems and challenges that stand between using autonomous agents with their full power, while doing it in a way that’s safe. 

Building for the enterprise

“In the next few months, we are looking at large business enterprises that have strict security requirements, compliance requirements, regulatory requirements, etc. So we’ve already gotten to the point where we have the offering where you can connect to your email, you can connect to your calendar and do that safely where any sensitive action has approval.”

Obviously, entry to enterprise space will be a challenge for any small operation. But NanoClaw knows what it is reaching for, so they could be the right crustacean to back as the competitors fight for OpenClaw’s dropped crown.

The post Gavriel Cohen found his own code inside OpenClaw, so he walked away appeared first on The New Stack.