From AI Assistants to Acting Enterprise AI Agents
AI in business is shifting from passive copilots to autonomous enterprise AI agents that execute tasks, call tools and run continuously across core processes. At SAP Sapphire, this transition took center stage as SAP positioned agentic AI as the next productivity wave for finance, procurement, supply chain and manufacturing. Yet the move from generating answers to taking actions fundamentally changes risk: agents can now touch systems of record, traverse application and data boundaries and operate without human oversight at every step. Traditional chatbot-era safeguards are no longer enough. Enterprises need agents that are safe, governable and auditable by design before they can leave the lab and enter production. The expanded SAP NVIDIA partnership directly targets this gap, reframing enterprise AI not as a loose collection of models, but as a tightly governed execution fabric for specialized agents.
Inside the SAP NVIDIA Partnership on Agent Governance
Announced during SAP CEO Christian Klein’s keynote, joined via video by NVIDIA founder and CEO Jensen Huang, the SAP NVIDIA partnership centers on co-defining how enterprise AI agents are built, governed and executed. SAP is embedding NVIDIA OpenShell—an open source, secure runtime for autonomous AI agents—into the SAP Business AI Platform as the default runtime security layer. This applies not only to SAP-delivered agents but also to custom agents designed in Joule Studio, SAP’s environment for building end-to-end enterprise AI agents. Rather than simply adopting an external runtime, SAP is codesigning OpenShell alongside NVIDIA, contributing engineering resources to harden it for mission-critical workloads. The shared objective is clear: give enterprises a consistent way to run specialized agents with strong security and governance controls, so organizations can move confidently from pilots to production deployments.
OpenShell: Secure AI Execution as the Foundation
NVIDIA OpenShell provides the low-level safeguards that make secure AI execution possible for autonomous agents. As an open source runtime, it creates isolated execution environments where each agent operates within tightly defined boundaries. Policy enforcement spans filesystem and network access, ensuring that agents see only what they are allowed to see and can interact only with approved resources. Crucially, OpenShell adds infrastructure-level containment that limits the blast radius if agent logic fails or behaves unexpectedly, preventing errors from cascading into critical systems. This secure runtime layer effectively asks a single question before any action executes: can this agent action safely run? By placing OpenShell at the heart of SAP’s Business AI Platform, the partnership establishes a standardized, auditable execution substrate that enterprises can rely on across varied workloads and regulatory environments.
SAP’s Enterprise Control Layer: From Safety to Should This Happen?
While OpenShell focuses on whether an action can execute safely, SAP’s Joule Studio runtime addresses the higher-order question: should this action happen at all? Operating as the enterprise control layer in SAP Business AI Platform, Joule Studio brings business-aware semantics to agent governance—roles, skills, lifecycle states, permissions and process contexts. It aligns agent behavior with enterprise identity and access management, compliance frameworks and policy rules that security and audit teams already understand. SAP is feeding real-world agentic workloads and requirements into OpenShell’s evolution, shaping isolation boundaries, policy models and auditability to match enterprise risk profiles. Together, Joule Studio’s governance controls and OpenShell’s secure runtime close a critical gap that application-layer security alone cannot, enabling AI agents to operate within the same policy, identity and process constraints that govern traditional SAP applications.
Why This Collaboration Changes the Enterprise AI Playbook
The SAP NVIDIA collaboration reframes how enterprises think about deploying specialized AI agents at scale. Instead of bolting security on after experimentation, governance and safety are embedded in the execution stack from day one. Enterprises gain clear audit trails for agent actions, runtime-level containment for failures and policy enforcement aligned with existing IAM and compliance structures. NVIDIA also brings its own perspective as a large SAP customer, grounding the design of OpenShell in real operational demands. For organizations running mission-critical workloads, this means they no longer face a trade-off between innovation and control. The combined stack—OpenShell as the secure execution layer and Joule Studio as the enterprise control plane—sets a reference architecture for trustworthy enterprise AI agents, accelerating adoption while meeting the scrutiny of security, risk and regulatory teams.