SAP launches AI Agent Hub at Sapphire 2026 to tame vendor agent sprawl

SAP wants to be the place where enterprises manage every AI agent in their environment, no matter which vendor built it. 

At Sapphire 2026 in Orlando on Tuesday, the company opened up the SAP AI Agent Hub, its vendor-agnostic command center for inventorying and governing AI agents, LLMs, and Model Context Protocol (MCP) servers, to more SAP customers through Joule Studio.

Agent Hub was previously available only to customers of SAP LeanIX, the company’s product that helps enterprises visualize and analyze their IT landscape. In its new form, the hub now covers every agent, large language model (LLM), and MCP server in an enterprise, regardless of who built it and where it was built. Two of its six capabilities are generally available today; four are scheduled for the third quarter of 2026.

It seems inevitable that every large enterprise will eventually end up with hundreds of agents (or more) from many vendors, such as Microsoft Copilot and Salesforce Agentforce, as well as AI-native agents from Anthropic and OpenAI, custom-built agents from LangGraph or AutoGen, and SAP’s own Joule Agents. As of now, they typically sit in different systems with no central inventory and no audit trail for IT and security teams.

“If I put myself in the shoes of every IT department, it’s great to build agents, but you want to have control.”

SAP Business Technology Platform president Michael Ameling tells The New Stack, “If I put myself in the shoes of every IT department, it’s great to build agents, but you want to have control. Otherwise, you basically run into this challenge everybody had at the beginning of the web services era.”

The core idea behind the SAP AI Agent Hub is to avoid unmanaged agent sprawl and provide enterprises with a single system of record for all their AI assets.

What the hub does

One of the core features of such a system is, unsurprisingly, a registry. The AI registry in Agent Hub is now generally available. 

To be useful, any registry has to make it as easy as possible to create an authoritative index. To do this, the Agent Hub auto-discovers agents, LLMs, and MCP servers across vendors. This registry is now generally available.

But building an inventory is just one of the things the Agent Hub does. It also provides tools to evaluate and verify workflows that capture risk ratings and compliance mappings for each agent. This ensures that nothing ships into production without a verified governance record. Identity and access control, coming in Q3 2026, give each agent a unique identity through SAP Cloud Identity Services.

“We give each agent a unique identity by default… you want to authorize, give them access control — also to data and the like — and have auditability.”

“We give each agent a unique identity by default,” Ameling says. “It’s very important: you want to authorize, give them access control — also to data and the like — and have auditability.”

AI observability will also be part of the hub and is planned for Q3 2026. This adds the kind of session-level telemetry (health, tool-call correctness, root-cause analysis, etc.) that companies like LangSmith and Datadog already do for agents you’ve instrumented yourself. 

“You can see who is interacting with whom, what the behavior of these agents is,” Ameling explains. “Do I actually use all tools? Or maybe in 50 percent of the cases, a human-in-the-loop interaction is required. Maybe it’s not that efficient. Maybe you should have a look at how you can optimize your agent.” 

Agent mining gets practical

Maybe the most interesting piece here is agent mining. SAP Signavio is applying process mining to AI agents. Originally built to spot gaps between designed and actual business workflows, this tool is now used to determine whether an agent is following the execution pathway it was designed for. Agents are, after all, non-deterministic by default. 

SAP’s moat

SAP’s advantage in this space is that it starts with a stack that already maps the architecture (LeanIX), processes (Signavio), identities (Cloud Identity Services), and the org chart (SuccessFactors).

SAP isn’t totally new to this. In November 2025, SAP Signavio framed agent mining as one of four pillars of an “AI agent excellence” approach: discovery, context, mining, and value impact. Now, those pillars are part of a single Hub umbrella in Joule Studio, rather than scattered across Signavio and LeanIX.

It’s worth noting that SAP’s competitors are building towards a similar goal, but from different starting points. Microsoft is stitching Copilot Studio together with its Entra and Purview services, while AI-native observability vendors like LangSmith also cover parts of the picture. But for these companies, the architecture, identity, and HR components native to SAP may be harder to add on.

Before joining The New Stack as its senior editor for AI, Frederic was the enterprise editor at TechCrunch, where he covered everything from the rise of the cloud and the earliest days of Kubernetes to the advent of quantum computing….

Read more from Frederic Lardinois