Enterprise AI has moved beyond generating text and reasoning through problems. Now, companies are asking a more pressing question: How should AI act? Early autonomous agents showed promise, but deploying them in enterprise environments requires context, control, and consistency across real workflows. At ServiceNow Knowledge 2026, NVIDIA founder Jensen Huang and ServiceNow CEO Bill McDermott jointly announced a powerful expansion of their collaboration—introducing specialized, safe, and scalable autonomous AI agents. This partnership combines NVIDIA's accelerated computing, open models, and secure agent execution with ServiceNow's Action Fabric and AI Control Tower. Here are seven transformative aspects of this new initiative that every enterprise should understand.
1. A Visionary Partnership at Knowledge 2026
During the opening keynote, Jensen Huang and Bill McDermott laid out a bold vision for enterprise AI—one where agents don't just respond to prompts but actively manage complex, multi-step tasks. The two companies are expanding their collaboration across the full technology stack, from hardware to software to governance. This isn't a simple integration; it's a deep co-engineering effort aimed at delivering agents that are both powerful and enterprise-safe. By leveraging NVIDIA's cutting-edge accelerated computing and open-source foundations, paired with ServiceNow's workflow expertise, the duo is setting a new standard for AI autonomy in business. This partnership signals that the era of passive AI is giving way to proactive, autonomous agents that can transform how work gets done.
2. Introducing Project Arc: The Self-Evolving Desktop Agent
Project Arc is a standout innovation—a long-running, self-evolving autonomous desktop agent designed specifically for knowledge workers such as developers, IT teams, and administrators. Unlike standalone agents that operate in isolation, Project Arc connects natively to the ServiceNow AI Platform through the Action Fabric. This connection brings governance, auditability, and workflow intelligence to every action the agent takes. It can access local file systems, terminals, and installed applications to complete tasks that traditional automation tools cannot handle—such as orchestrating complex deployments or troubleshooting across multiple systems. Crucially, it does so with the controls enterprises need to deploy AI at scale, ensuring every action is logged, auditable, and reversible.
3. Secure Agent Execution with NVIDIA OpenShell
Security is non-negotiable for enterprise AI, and Project Arc addresses this head-on using NVIDIA OpenShell, an open-source secure runtime environment. OpenShell allows developers to create and run autonomous agents in sandboxed, policy-governed settings. Enterprises can define exactly what an agent can see, which tools it can use, and how each action is contained—preventing sensitive data exposure or system damage. ServiceNow is not just using OpenShell; it is actively contributing to its development, helping to establish a common foundation for secure agent execution across the industry. This runtime layer ensures that autonomous agents operate within strict guardrails, giving businesses the confidence to deploy them widely.
4. Governance Through ServiceNow AI Control Tower
Even the most capable agent needs oversight, which is where the ServiceNow AI Control Tower comes in. This governance layer provides centralized visibility and control over all AI agents, including Project Arc. It monitors behavior, enforces policies, and maintains audit trails for every action taken by an agent. Combined with the secure runtime of OpenShell, the Control Tower offers a dual-layer approach to safety: the runtime restricts what an agent can do, and the Control Tower ensures that those restrictions are enforced and recorded. This gives IT leaders the confidence to let agents operate autonomously, knowing that any deviation can be quickly detected and corrected. It's a critical component for gaining regulatory compliance and internal trust.
5. Action Fabric: Connecting Agents to Real Workflows
The ServiceNow Action Fabric is the connective tissue that makes Project Arc truly enterprise-ready. It bridges the gap between the autonomous agent and the organization's existing workflows, data sources, and applications. Instead of building custom integrations, enterprises can use the Action Fabric to grant agents context-aware access to the tools they need. For example, an agent could pull incident data from IT service management, check a knowledge base, and update a ticket—all within a governed framework. This integration ensures that agents operate with the same intelligence and context that human workers rely on, making their actions relevant and accurate. The Action Fabric also feeds back into the AI Control Tower, enabling continuous improvement and monitoring.
6. Open Models and Custom Skills for True Adaptability
Enterprise AI cannot be one-size-fits-all. Recognizing this, NVIDIA and ServiceNow are emphasizing open models and domain-specific skills. Project Arc is built on a foundation that allows customization: companies can fine-tune the underlying language models to their unique domain—whether finance, healthcare, or manufacturing—and deploy specialized skills that the agent can use autonomously. This openness means enterprises are not locked into a single vendor's ecosystem. They can choose from a range of open-source models from NVIDIA's catalog and add custom skill modules that match their business processes. The result is an agent that understands industry jargon, internal procedures, and regulatory requirements, making it far more effective than a generic chatbot.
7. Three Pillars for Autonomous Enterprise AI
The entire Project Arc design is based on three essential requirements that every enterprise will need for long-running, autonomous agents. First, open models and domain-specific skills that can be customized and fine-tuned. Second, security mechanisms that help agents act without exposing sensitive data or systems—this is where OpenShell and the Control Tower shine. Third, all of this runs on AI factories that deliver efficient tokenomics, meaning the computational cost of running these agents is optimized for scale. These three pillars—adaptability, security, and cost-efficiency—form the backbone of a truly autonomous enterprise AI strategy. Without any one of them, agents risk being too rigid, too risky, or too expensive to deploy broadly.
The collaboration between NVIDIA and ServiceNow marks a decisive step forward in enterprise AI. By combining cutting-edge hardware, open-source security, and robust workflow integration, they have created a blueprint for autonomous agents that businesses can trust. Project Arc is not just a product—it's a paradigm shift. As enterprises continue to explore what AI can do when it is allowed to act independently, these seven elements will serve as critical guideposts. The future of work is autonomous, and with the right foundations in place, it promises to be both powerful and safe.