SAP's AI Agent Hub: when agent sprawl hits the boardroom, governance becomes a product

SAP positioned AI Agent Hub as a governance product just as agent sprawl became a board-level issue, but only two of its six planned capabilities, the agent registry and the verification workflow, are generally available today. The verification badge gates by reference, not by inline runtime enforcement, and independent scoring (Gartner's first AI Governance Magic Quadrant, a Kosmoy comparison) puts SAP's runtime and security axes well behind its inventory strength. For S/4HANA teams, the useful frame is composability rather than a binary build-vs-buy choice: take the inventory and business-context layer, especially if LeanIX is already licensed, and source runtime enforcement separately. A BTP routing requirement for agentic API access adds a second, cost-side decision on top of the governance one.
Every enterprise architecture team I talk to this quarter is wrestling with the same quiet problem: nobody can produce a clean inventory of which AI agents are actually running against their systems. SAP's own numbers, published alongside its AI Agent Hub rollout, put a figure on that discomfort. In the SAP LeanIX Agentic AI Survey 2026, 98 percent of companies said they had already deployed AI agents or planned to, while less than half reported having visibility into an inventory of those agents. Gartner's read is blunter still: the average Fortune 500 enterprise is projected to run more than 150,000 AI agents by 2028, up from fewer than 15 in 2025, and only 13 percent of organizations currently believe they have the right governance in place to manage that growth.
That gap is why SAP positioned AI Agent Hub, widened to all customers via Joule Studio at Sapphire 2026, as a governance product rather than a feature bolted onto an existing module. The pitch, in SAP CTO Philipp Herzig's own framing, is filling a "no central governance, no clear view" hole. For S/4HANA teams weighing whether to build their own agent oversight or buy into SAP's version, the honest answer depends on reading past the announcement to what is actually shipping, what is still a roadmap promise, and what an independent capability comparison says SAP is not yet built to do.
What is actually live, and what is still a promise

SAP's Hub covers six planned capabilities, but only two are generally available as of the Sapphire 2026 announcement: an AI registry that auto-discovers agents, LLMs, and MCP servers across vendors, and an evaluation and verification workflow that scores them. The other four; identity and access control through SAP Cloud Identity Services, AI observability covering session health and tool-call correctness, agent mining built on SAP Signavio process data, and org-chart and skills mapping via SuccessFactors, are scheduled for Q3 2026 according to SAP's own disclosure, reported by The New Stack. That date has not yet been independently confirmed as hit.
The registry piece is not new; it reached an earlier, narrower GA in November 2025 bundled free into SAP LeanIX Application Portfolio Management for existing LeanIX customers. The Sapphire 2026 move widened access beyond that base, but the underlying capability is the same inventory function SAP has offered for roughly a year. What is genuinely new in the 2026 version is the vendor-agnostic framing: the Hub indexes agents built on Microsoft, Google, AWS, Databricks, and ServiceNow stacks, not only SAP-built ones, according to SAP's own product page and Q2 2026 release notes.
A verification badge is not a runtime brake

The core governance mechanism in the Hub is what SAP calls a verification badge. Agents and MCP servers get evaluated, producing a risk rating and a compliance mapping, and integrated runtimes are meant to check that badge before allowing execution. An independent vendor comparison from Kosmoy describes this accurately as gating by reference, not inline runtime enforcement owned by SAP itself. In practice, that means the badge is only as strong as the runtime systems that choose to honor it. Nothing in the public material names a production deployment where an unverified agent was actually refused execution by a live system -- the mechanism is described as structural gating, with no cited customer actively enforcing it today.
That distinction matters for an S/4HANA estate deciding where to put its governance budget. A registry tells you what exists. A policy gateway, identity binding, and containment decide what is allowed to run and what happens when it misbehaves. SAP's Hub is strong on the first and, as of this GA date, thin on the second.
Where the independent scoring gets uncomfortable
Gartner published its first Magic Quadrant for AI Governance Platforms in June 2026, covering 13 vendors. SAP landed in the Niche Players quadrant, alongside Relyance AI, Cranium AI, and Saidot; ServiceNow, in the same MQ, was placed as a Leader. Three independent secondary analyses corroborate that placement.
A separate, more granular comparison from Kosmoy scores ten vendors across capability axes on a 0 to 10 scale. SAP's registry and business-context strength score well (AI inventory and discovery at 7, compliance and audit at 6, agent building via Joule Studio at 6), but the runtime and security axes are where the gap opens: security and shadow AI at 4, gateway and policy control at 3, observability and FinOps at 2, guardrails and runtime safety at 2, agent containment at 2, and testing, evals, and red-teaming at 1. Microsoft Agent 365, which went GA on May 1, 2026 with native Entra Agent ID identities and conditional access by default, scores 7 to 9 across most of those same security and runtime axes.
Kosmoy is a competing AI-governance vendor, which is a real conflict-of-interest risk worth naming rather than ignoring. The mitigating fact is that Kosmoy's specific GA-date claims about SAP (two of six capabilities shipped, four on a Q3 2026 roadmap) match SAP's own sponsored coverage in The New Stack almost exactly. The facts are corroborated by a second, differently biased source; it is the comparative scoring and ranking, not the underlying feature inventory, that should be read with Kosmoy's commercial position in mind.
The build-vs-buy math for S/4HANA teams
For an architecture team already running LeanIX, the registry and verification workflow come essentially free, and that is a genuine starting point: business-context grounding drawn from LeanIX architecture data, Signavio process mining, and SuccessFactors org mapping is not trivial to replicate in a homegrown tool. A production case study from SAP-funded integrator DEBCOR Engineering frames the Hub as layer five of a five-layer agent architecture (worker, intelligence, orchestration, governance, auditing), used for segregation-of-duties checks and action-scope authorization against live SAP FI data. Notably, even that architecture pairs the Hub with LangGraph and n8n for orchestration and Anthropic's Claude for reasoning; no single vendor, including SAP, is supplying the full governance stack in that deployment.
That composability is probably the more useful frame than a binary build-vs-buy choice. A team whose risk model requires inline blocking, not just detection and badging, needs a runtime-security layer regardless of whether SAP's Hub sits above it: Microsoft Agent 365, ServiceNow AI Control Tower, Zenity, Noma, or Cisco AI Defense are the names showing up in that conversation today. The Hub's job in that stack is inventory and business-context mapping, done early and done free if you already hold a LeanIX license. The runtime enforcement job stays open.
There is also a cost wrinkle specific to S/4HANA estates that sits adjacent to the governance question but shapes it anyway. Independent analyst Rajesh Beri has reported that SAP's API usage policy now requires agentic access to S/4HANA OData APIs to route through SAP BTP or SAP Business Data Cloud rather than calling S/4HANA directly, regardless of which governance layer, SAP's own or a third party's, sits on top. Any build-vs-buy decision on agent governance for an S/4HANA shop is really two decisions stacked together: who governs the agents, and who gets paid for the API path they run on.
Reading the hype against the gap, fairly
None of this means SAP is manufacturing a problem to sell a solution. The Gartner and SAP LeanIX survey numbers describe a real, broad visibility gap that predates any vendor's product launch, and a first-year Niche Player placement in a brand-new Magic Quadrant category is a meaningfully different signal than a mature product falling behind in an established one. Gartner's methodology rewards completeness of vision and ability to execute at a specific point in time; SAP is competing in a category that did not formally exist in a Gartner document before June 2026.
The practical question for an enterprise architecture lead evaluating this today is narrower than "is SAP's Hub good or bad." It is whether the governance architecture question (who has an inventory, who owns business context) and the runtime enforcement question (what actually gets blocked, and by what) are being answered by the same purchase, or whether the inventory piece is genuinely free and worth taking now, while the enforcement piece stays an open, separately sourced decision until SAP's Q3 2026 roadmap items ship and get independently verified against a production system.
Sources
- SAP News Center, "AI Agent Sprawl: Why AI Governance Is Now a Board-Level Issue," Paul Taylor, Aug 3 2026. https://news.sap.com/2026/08/agent-sprawl-why-ai-governance-is-now-board-level-issue/
- SAP product page, "SAP AI Agent Hub: AI Governance and Agent Management System." https://www.sap.com/products/artificial-intelligence/ai-agent-hub.html
- The New Stack, "SAP launches AI Agent Hub at Sapphire 2026 to tame vendor agent sprawl," Frederic Lardinois. https://thenewstack.io/sap-ai-agent-hub/
- Gartner newsroom, "Gartner Identifies Six Steps to Manage AI Agent Sprawl," Apr 28 2026. https://www.gartner.com/en/newsroom/press-releases/2026-04-28-gartner-identifies-six-steps-to-manage-artificial-intelligence-agent-sprawl
- Kosmoy, "Best AI Agent Governance Platforms in 2026: 10 Compared," last verified Aug 17 2026. https://www.kosmoy.com/resources/blog/best-ai-agent-governance-platforms-2026/
- Kosmoy, "Best AI Governance Platforms in 2026: 14 Enterprise Vendors Compared." https://www.kosmoy.com/resources/blog/best-ai-governance-platforms-2026/
- Medium (sanjmo), "Inside Gartner's First AI Governance Platform Magic Quadrant," Jul 10 2026. https://sanjmo.medium.com/inside-gartners-first-ai-governance-platform-magic-quadrant-fa1f182f75e9
- SAP Community Enterprise Architecture blog, "AI Governance and Cost Monitoring with SAP AI Agent Hub and SAP for Me," Amit_SAP_UKI. https://community.sap.com/t5/enterprise-architecture-blog-posts/ai-governance-and-cost-monitoring-with-sap-ai-agent-hub-and-sap-for-me/ba-p/14507505
- DEBCOR Engineering, "SAP AI Agents Built and Running in Production." https://www.debcor.com/ai/sap-agents
- beri.net (Rajesh Beri), "SAP API Policy Prohibits Agentic AI Unless You Pay for BTP," 2026. https://www.beri.net/article/sap-api-policy-ai-governance-2026
- Microsoft Security Blog, "Microsoft Agent 365 now generally available," May 1 2026. https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/
Frequently asked
Which SAP AI Agent Hub capabilities are generally available?
Two of the six planned capabilities are generally available: an AI registry that auto-discovers agents, LLMs, and MCP servers across vendors, and an evaluation and verification workflow that scores them. The other four (identity and access control, AI observability, agent mining on Signavio data, and org-chart and skills mapping via SuccessFactors) are scheduled for Q3 2026 per SAP's own disclosure, and that date has not been independently confirmed as hit.
Does the SAP AI Agent Hub verification badge block unverified agents at runtime?
Not by itself. Agents and MCP servers are evaluated and receive a risk rating and compliance mapping, and integrated runtimes are meant to check the badge before allowing execution. That is gating by reference, not inline enforcement owned by SAP, so the badge is only as strong as the runtime systems that honor it. No public material names a production deployment where an unverified agent was actually refused.
Should S/4HANA teams build or buy agent governance?
Treat it as composable rather than binary. For teams already running LeanIX, the registry and verification workflow come essentially free and are hard to replicate in a homegrown tool. Teams whose risk model needs inline blocking still need a separate runtime-security layer, and the DEBCOR production case pairs the Hub with LangGraph, n8n, and Claude rather than relying on SAP for the full stack.
Need this in your organisation?
I work with a small number of clients each quarter on ERP strategy and IT-department automation. If the questions raised above are live in your team, get in touch.
Start a conversation →