You’re reading an issue of “The AI Economy,” my newsletter exploring the forces shaping the AI era—tracking how AI is rewriting business, work, technology, and culture. Subscribe to get expert insights and curated updates delivered straight to your inbox.
For years, enterprises have struggled to govern AI where it actually runs, across clouds, tools, and now, increasingly, the desktops where employees do their work. ServiceNow thinks it has an answer, and it’s building the solution with Nvidia. Project Arc is the result: an enterprise desktop agent designed to bring autonomous capabilities seen in tools like OpenClaw into the corporate space, but with the security guardrails and governance that IT leaders actually require.
Announced at ServiceNow’s Knowledge 2026 conference, Project Arc is powered by Nvidia’s OpenShell, a sandbox environment where “everything is denied by default,” and IT administrators can define policies on what’s allowed. Installed on an employee’s desktop or laptop, this container hosts the autonomous agent, which has access to whatever the employee grants it and can take actions on their behalf. Everything is overseen and managed by ServiceNow’s AI Control Tower, while also grounded in the company’s Configuration Management Database (CMDB).
“Long-running, autonomous agents are rapidly changing the game for enterprise AI, and delivering them securely at scale requires governance that spans models, software, and AI infrastructure,” Kari Briski, Nvidia’s vice president of generative AI for enterprise, said in a statement.
Employees can think of Project Arc as similar to AI-powered IT support for handling mundane, routine tasks. The autonomous agent is capable of thinking, code writing, executing tasks, and adapting when things don’t go as planned. ServiceNow claims it can handle multi-step work across enterprise tools and systems without requiring prebuilt workflows.
“What we’re trying to help customers with ultimately is automation,” Joe Davis, ServiceNow’s executive vice president of AI engineering and delivery, tells The AI Economy in an interview.
Project Arc also works in tandem with ServiceNow’s AI specialists, prebuilt agents designed for specific business domains like IT and HR. These agents handle cloud-based requests by default—a Zoom access request, for instance, is really an entitlement issue with nothing to do with the desktop. But when a task requires direct access to an employee’s machine, such as restarting a browser or creating an email cache, Davis said that’s where Project Arc steps in. “That’s where Arc can come in because it can take action on a user’s machine.”
He acknowledges that IT-specific desktop agents could be next on ServiceNow’s roadmap. “There [are] a lot of settings and applications that need to be configured or changed to fix things—you’ve got to change some settings and restart,” Davis said. Some self-evolving desktop agent will be really good at that, and it will teach itself how to fix these things, and then we can share that across users.”
“The other thing that we’re looking at here is it’s pretty hard to set this stuff up today,” he went on to say. “There’s a lot of software that you have to grind through to get these working. We think that we can make this turnkey and really easy for enterprises to deploy safely.”
To be clear, Project Arc isn’t ServiceNow’s answer to Microsoft’s Copilot+ PC. The goal isn’t local AI processing—most tasks will run in the cloud, with sensitive work handled on-device when possible. ServiceNow isn’t asking companies to refresh their hardware either, though Davis acknowledged a practical tradeoff: older machines will lean heavily on the cloud, and organizations “have to be comfortable with what’s flowing back and forth.”
This isn’t new territory for ServiceNow. The company already runs robotic process automation agents that handle tasks across Windows applications and web agents that operate inside browsers. But the competitive landscape is intensifying. Last year, Microsoft upgraded its Edge browser to let developers integrate AI features directly into web apps, and the company is also exploring OpenClaw technology for its own Microsoft 365 Copilot ambitions. Atlassian acquired the startup behind Dia, an AI-powered browser, last September. And AI model makers—Anthropic, OpenAI, and Perplexity among them—have staked their own claims in the space.
ServiceNow isn’t backing down. “There’s a lot of experimentation that customers and companies want to do right now, so I don’t think there’s a winner-take-all in the AI space anytime soon,” Davis remarked. “We’re seeing companies that want to place multiple bets and see who wins and what happens in each of these categories.”
Widespread enterprise adoption of autonomous desktop agents is still nascent, but the window may be closing faster than expected. The emergence of OpenClaw and others in the so-called “Claw” family, including Nvidia’s own NemoClaw, has reset expectations for what these agents can do and how quickly they can do it.
The harder challenge isn’t the technology. It’s compliance. Davis acknowledged that heavily regulated industries, like financial services and pharmaceuticals, face a longer road, noting that “there’s going to be a number of customers we have to work with to get through some of these things.” ServiceNow sees AI Control Tower as its advantage here, a cross-platform governance layer that abstracts away the underlying operating system, with Nvidia’s partnership helping solve the trickier problem of edge connectivity to local devices.
Project Arc is currently available as an early preview. There is no timeline for when it’ll publicly launch.
Disclosure: I attended ServiceNow’s Knowledge 26 as a guest of the company, with my flights and hotel stay paid for. The AI Economy’s coverage is editorially independent from those that it covers. These words are my own.
