AI Agents are no longer waiting for instructions - they're acting. ServiceNow Cowork (formerly announced as Project Arc at K26), puts an autonomous AI teammate directly on the employee desktop, where it thinks, writes code, calls APIs, and completes complex multi-step work across enterprise systems without requiring pre-built workflows. Because of its local-first execution, it can reach the work the employees actually do, not just the systems a central agent has been wired into. Available today as a sneak preview, ServiceNow Cowork represents ServiceNow's boldest step yet toward making autonomous work a daily enterprise reality. But autonomy without control is a liability. That's where NVIDIA OpenShell comes in — a sandboxed runtime that wraps every single action the agent takes, ensuring nothing runs outside a contained, policy-governed environment. And sitting above it all, ServiceNow AI Control Tower sets the policies, monitors behavior in real time, and govern the fleet of ServiceNow Cowork deployed in the enterprise. In this session, you'll see what that looks like in practice: an agent resolving incidents autonomously, then the instant it reaches outside its boundaries, it's blocked with the audit trail to prove it. This is not a prototype. This is enterprise AI that is secure, auditable, and production-ready from day one.