NVIDIA and ServiceNow Launch Autonomous AI Agent Era
Enterprise AI has evolved rapidly from simple text generation to advanced reasoning. At the ServiceNow Knowledge 2026 conference, NVIDIA CEO Jensen Huang and ServiceNow CEO Bill McDermott announced the next major leap: autonomous AI agents designed to act on behalf of enterprise employees. This expanded collaboration aims to deliver specialized, governed agents that can execute complex tasks from the desktop to the AI factory.
Project Arc: The Autonomous Desktop Agent
Central to this announcement is Project Arc, a self-evolving autonomous desktop agent built specifically for knowledge workers, developers, and IT administrators. Unlike standalone bots, Project Arc is natively integrated into the ServiceNow Action Fabric. This connection ensures that every action taken by the agent is governed, auditable, and aligned with enterprise workflow intelligence.
Project Arc can navigate local file systems, use terminal applications, and interact with installed software to complete multi-step tasks that were previously impossible to automate. To ensure security, these agents operate within NVIDIA OpenShell, an open-source secure runtime that sandboxes agent activities according to strict policy-governed environments.
Specialized Skills and Open Models
To make these agents effective across different industries, NVIDIA and ServiceNow are prioritizing an open ecosystem. Key technical components include:
- NVIDIA Agent Toolkit: Provides flexible building blocks and Nemotron open models for developing customized applications.
- NVIDIA AI-Q Blueprint: A specialized framework for building deep research agents capable of synthesizing vast amounts of information to support decision-making.
- NOWAI-Bench: A new open benchmarking suite, integrated with NVIDIA NeMo Gym, designed to evaluate how agents handle complex, real-world enterprise workflows.
In these evaluations, the Nemotron 3 Super model currently ranks as the top-performing open-source model for challenging enterprise tasks.
Scaling with Efficient AI Factories
As agents transition to "always-on" status, the cost of generating tokens becomes a critical business factor. NVIDIA's AI Factory infrastructure, specifically the Blackwell platform, is designed to address this by delivering 50x greater token output per watt compared to the previous Hopper architecture. This efficiency results in a nearly 35x reduction in cost per million tokens, allowing enterprises to scale Agentic AI across millions of internal workflows without prohibitive expenses.
