Ask anyone managing enterprise software right now how many AI agents are actually running in their environment, and you will get a version of the same answer: they do not know. Shadow IT has a new, higher-risk cousin.
SAP is betting its installed base of roughly 425,000 customers can solve this. The AI Agent Hub, built on SAP LeanIX, is a vendor-agnostic command center for discovering, inventorying, and governing AI agents across the enterprise—SAP and non-SAP agents, LLMs, and Model Context Protocol servers included. It was announced in May 2026 at SAP Sapphire, not this month. The September activity—a pragmatic roadmap blog published September 22 and the SAP Transformation Excellence Summit in Atlanta that same week—is marketing reinforcement ahead of a Q3 2026 general availability.
What makes this worth watching is the distribution math. SAP is the first ERP-scale vendor to ship a dedicated agent inventory product. The company claims 150 to 180 organizations are already using the hub pre-GA, with over 180,000 agents discovered in production. Those numbers are entirely SAP-sourced—there is no independent audit—but the scale is suggestive enough that the claim is not trivial to dismiss.
The timing matters because the governance gap is getting expensive. IDC and Lenovo peg the pilot-to-production failure rate for custom agent builds at 88%. Only 23% of merchants can track agent-driven purchases, according to PYMNTS in September. If you cannot see what is running, you cannot measure it, govern it, or decommission it when it breaks.
SAP is not alone in chasing this problem. Dataiku shipped a standalone Agent Management product for cross-platform monitoring. NiCE paid $955 million for Cognigy’s routing layer. Collibra launched Guardian Agents for runtime governance. Okta moved into agent identity. Five governance-focused products shipped in roughly twelve days this month. The control layer is forming, and SAP’s entry arrives with the largest distribution channel in enterprise software.
The hub’s actual capabilities span discovery and inventory of all agents (SAP and non-SAP), post-deployment observability through SAP Cloud ALM (session tracing, goal completion monitoring), and agent behavior mining through SAP Signavio’s AI Agent Excellence framework—analyzing whether agents are deviating from intended workflows. SAP Company Memory, a governed knowledge layer for grounding agents in enterprise context, is in beta. The partnerships reach across Microsoft, Google Cloud, AWS, Anthropic, and NVIDIA, with bidirectional agent interoperability between Joule and external frameworks.
One important caveat: the claim that this is vendor-agnostic remains untested in highly heterogeneous, non-SAP-centric environments. Inventorying agents inside a clean SAP ecosystem is a different problem than doing it across a messy, multi-cloud architecture where different teams spun up different agent frameworks on different clouds at different times. That is where most enterprises actually live.
For the people running these systems, the value of the hub—if it works as advertised—is straightforward: you finally get to see what you deployed. Whether that visibility translates into actual governance, or just a shinier dashboard, depends on adoption well beyond SAP’s existing customer relationships. View this as a promising, distribution-backed attempt to bring some discipline to the agent-driven enterprise, not as a solved problem.