I am still reveling in the excitement of SAP Sapphire Orlando, May 11 to May 13, 2026.  It started with the SAP keynote , highlighting our future direction.  A personal moment of pride filled the screens around minute 42, when the CEO of NVIDIA Jensen Huang introduced how SAP and NVIDIA are making enterprise software more powerful and security.

 On Wednesday, I hosted a strategy talk between Sebastian Mahr, Chief Development Architect, SAP SE and Shahriar Hooshmand, GenAI Technical Lead, NVIDIA.  Today, I am sharing more details about their discussion for redefining agent execution. The AI first experience for building happens in Joule Studio and Joule Studio runtime with NVIDIA makes it easy to deploy and secure for enterprise scale.

AI agents are moving from conversation to action. They are beginning to execute tasks, invoke tools, cross system boundaries, and operate inside the business processes where decisions are made and value is created. For enterprises, this changes the trust equation. A chatbot can suggest; an agent can act. And once agents can engage with systems of record across finance, procurement, supply chain, manufacturing, and customer operations, the question becomes much sharper: how do we safely deploy autonomous AI? SAP and NVIDIA’s collaboration is designed to answer exactly that question. The SAP Business AI Platform uses the NVIDIA OpenShell secure runtime to make agents safe, governable, and auditable by design. Joule Studio acts as an “agent harness” that combines Large Language Model intelligence with business data and domain expertise.

Autonomy Intelligence, and Security

Every enterprise agent sits in a tension between Autonomy, Intelligence, and Security. Highly autonomous agents can move fast, intelligent agents can reason through complex business contexts, and secure agents can operate within clear technical and business boundaries. In real-world enterprise environments, you cannot simply maximize all three at once. A loan-processing agent, a production-line response agent, and a strategic planning agent should not run with the same level of freedom, the same approval model, or the same risk posture. The more autonomous the agent, the more important it becomes to define what it can see, what it can do, who approves its actions, and how those actions are traced.

Responsibility and secure execution must become the layer around agentic enterprise AI. Security is about container isolation, credentials, networks, APIs and risk management for unauthorized spending, compliance violations, reputational damage, liability, and audit readiness. Enterprises need a way to choose the right balance for each workload:

• Autonomous and secure for repeatable tasks,
• Intelligent and secure for high-context work with human oversight, or
• Autonomous and intelligent only in tightly controlled development and testing environments.

Responsibility defines who decides the operating mode, who bears the risk, and how the system adapts as the workload becomes more capable or more sensitive.

In Joule Studio runtime, NVIDIA OpenShell provides the secure runtime foundation for agent execution, including isolated environments, filesystem and network policy enforcement, and runtime-level containment to limit impact if agent logic fails. SAP brings the enterprise context: roles, skills, identity, lifecycle, policy semantics, observability, auditability, and governance across business landscapes. Put simply, OpenShell helps answer, “Can this agent action safely execute?” Joule Studio runtime helps answer, “Should this action happen at all?” Together, they close the gap between technical containment and business accountability.

The SAP and NVIDIA collaboration, which includes SAP as a key contributor to the NVIDIA OpenShell open source project, also reflects NVIDIA's broader full-stack view of AI. NVIDIA’s founder and CEO, Jensen Huang, has framed AI as a five-layer stack;  energy, chips, infrastructure, models, and applications; the application layer is where AI creates value in real workflows. For SAP customers, that point matters. Enterprise AI becomes real when agents can operate inside the business applications, identities, policies, and audit models that companies already depend on. SAP and NVIDIA’s work on Joule Studio runtime is about making that next step practical: allowing agents to act, while staying within the boundaries enterprises require.

This week at Computex / GTC Taipei, NVIDIA is announcing new open source software and models to continue making it easier to build and deploy autonomous agents. And now, the next step is to put this into the hands of builders. SAP customers and partners building custom agents should engage with the new Joule Studio early access / Early Adopter Program to start shaping their own trusted agent execution model. Explore how SAP enterprise with OpenShell-based runtime security and governance powered by OpenShell can help you understand your security posture, define workload-specific risk profiles, and move from pilots to production with confidence. Autonomous enterprise is no longer only a vision. With SAP and NVIDIA working together on secure, responsible agent execution, it is beginning now.

Now is the time to get in line.