SAP LeanIX was built to help companies make sense of sprawling tech stacks. Now SAP wants to apply much of the same logic to AI.
As companies deploy more AI agents, models, and MCP servers, keeping track of what is running across the organization is becoming its own architecture problem. An agent built by finance may connect to one model and a handful of internal systems, while engineering runs several others through a completely different stack.
Multiply that across business units and the inventory starts getting out of hand.
SAP is positioning LeanIX as one place to map that landscape.
Its AI Agent Hub is designed to catalog AI agents, models, and MCP servers and connect them with the apps, business capabilities, processes, and owners already represented in LeanIX.
To get a sense of the size of the challenge, the SAP LeanIX 2026 Agentic AI Survey offers a few useful numbers. The report found that 44% of companies were already using AI agents and another 40% were experimenting with them. Yet only 17% said they had visibility into agent performance or conformance, while 48% lacked clear roles and responsibilities for managing agents.
The more agents, models, and AI tools pile up, the harder it becomes to see what is actually there. It is almost as if the AI landscape is being covered by a thick layer of fog. SAP wants to clear some of that fog with LeanIX.
First, Where Did LeanIX Come From?
LeanIX has been around much longer than the current wave of AI agents.
The company was founded in Bonn, Germany, in 2012 by André Christ and Jörg G. Beyer. From the beginning, its focus was enterprise architecture management, giving companies a more usable way to understand the technology landscape they had accumulated over time.
Years of acquisitions, regional deployments, cloud migrations, departmental purchases, and one-off projects can leave you with hundreds or thousands of apps spread across the organization.
Some are critical. Others overlap. Some have barely any users left. A few may depend on technology approaching end of life. And often, nobody has a complete picture of how all those pieces connect.
That leads to two problems IT leaders know very well. SaaS sprawl is what happens when applications keep accumulating across the company, often with overlapping functions, rising costs, and unclear ownership. Shadow IT is closely related: employees or business units start using technology outside the visibility or approval of central IT.
LeanIX built its business around bringing some order to that mess. And in 2023, SAP decided it wanted that capability inside its own transformation portfolio, acquiring LeanIX and completing the deal in November of that year.
Where Does LeanIX Fit Inside SAP?
Today, SAP LeanIX sits inside SAP’s broader Business Transformation Management portfolio. It is still a standalone product, so you do not need another SAP solution to use it, but SAP increasingly positions it as part of a connected transformation toolchain.
The portfolio brings together three products with fairly different jobs.
- SAP LeanIX focuses on the enterprise architecture: the applications, technologies, dependencies, business capabilities, and increasingly the AI assets sitting across the company.
- SAP Signavio looks at the process side. It helps companies understand how business processes actually run, where they break down, and where they can be redesigned or standardized.
- And WalkMe sits closer to the user, helping companies understand how employees interact with technology and guiding them through new tools and processes as changes roll out.
You can think of them as three different views of the same transformation: LeanIX maps the technology, Signavio maps the processes, and WalkMe helps people adopt the changes.
So Where Does SAP AI Agent Hub Come In?
SAP AI Agent Hub extends that architecture view to AI agents, LLMs, and MCP servers.
SAP calls it a vendor-agnostic command center for discovering and governing AI assets across the business landscape. Its current discovery capabilities span assets coming from SAP as well as platforms from Microsoft, Google, AWS, Databricks, and ServiceNow.
For SAP LeanIX Application Portfolio Management customers, AI Agent Hub is available directly in the LeanIX workspace. SAP says this allows agents, models, and MCP servers to be governed in the context of the company’s full enterprise architecture.
Let’s see, in business terms, how specifically LeanIX do his job in mapping all the agents that are working in the company.
But what does that actually look like in business terms?
Let’s take an Invoice Exception Agent as an example.
An ordinary inventory could tell you that the agent exists. LeanIX goes a step further by connecting that agent to the rest of the company: who owns it, which business capability it supports, what model it uses, and which technologies and systems it depends on.
That becomes much more relevant when we’re talking about a few hundred agents.
With LeanIX, you may discover that finance and procurement have independently built agents doing similar work. Perhaps dozens of important agents depend on the same model provider. An MCP server that seemed relatively minor may turn out to be connected to several critical workflows.
Or maybe you find an agent that nobody can clearly identify an owner for, which is, of course, a pretty big red flag.
SAP explicitly points to these problems. Its documentation says independently developed agents can leave companies with duplicate agents, inconsistent controls, unclear ownership, and limited visibility into how AI interacts with applications, data, and processes.
The technology is new. The architecture problem is clearly familiar.
What Happens When the AI Stack Keeps Growing?
Seeing duplicate agents is useful. But the problem with AI sprawl goes way beyond than having a messy inventory.
The first issue is cost.
AI tools are increasingly being bought and consumed in different ways: per-seat subscriptions, usage-based pricing, token bundles, credits, and add-ons buried inside larger software contracts. And those costs do not always sit in one IT budget.
CIO recently reported that AI spending is increasingly hiding inside vendor renewals and individual business-unit budgets. Nearly two-thirds of companies surveyed said employees had used AI without proper oversight, while almost half of large enterprises did not have full visibility into the AI tools employees were using.
That makes duplication particularly easy.
And beyond the problem of tool duplication, there are other costs to consider. Some subscriptions may be heavily used. Others might keep renewing long after the original experiment ended.
The pattern is already familiar from SaaS. The difference with AI is that the bill can also keep moving after you buy the software, since many tools add consumption-based charges on top of licenses. Gartner recently called AI pricing uncertainty a material enterprise risk as vendors increasingly mix subscriptions, consumption, and outcome-based pricing.
Then there is the second problem: risk.
If IT does not know which AI tools are being used, it also becomes much harder to know which vendors have been reviewed, where company data is going, or what security controls those services have in place.
In May, Axios reported that security researchers had found around 380,000 publicly accessible assets created with AI coding platforms including Lovable, Base44, Replit, and Netlify. About 5,000 contained sensitive corporate information, including financial data and internal documents. Researchers said they encountered many of them while investigating shadow AI inside companies.
After all, if you don’t know if a tool is there, you probably have not evaluated the provider behind it either.
Is SAP AI Agent Hub More Than an Inventory?
Short answer: yes!
The current product goes well beyond keeping a list of agents. SAP AI Agent Hub covers much of the agent lifecycle, from planning and discovery to governance, monitoring, business-value analysis, and eventually decommissioning. Some capabilities are still under Early Adopter Care, so availability can vary.
SAP currently breaks that lifecycle into four broad stages:
1). Plan and Build. Before an agent goes live, teams can document architecture decisions, define ownership and governance criteria, and map the planned agent to the business capabilities and applications it will interact with. This is one of the places where LeanIX plays a particularly important role, because its enterprise architecture data gives teams the context to see what already exists, which systems the agent will touch, and even whether a similar agent or MCP server is already available.
2). Discover and Provision. Once agents start appearing across different platforms, Agent Hub can discover them and bring them into a common inventory for review. Within SAP LeanIX, those agents can then be linked to the applications, technologies, organizations, and business capabilities already mapped in the architecture.
3). Observe and Analyze. After deployment, SAP connects telemetry, usage, performance, verification status, and business-value information back to the agent portfolio. Other SAP products also enter the picture here. SAP, for example, points to SAP Cloud ALM for telemetry and SAP Signavio Process Intelligence for analyzing agents in the context of business processes.
4). Optimize and Decommission. Finally, companies can use that information to find agents that are underused, underperforming, or redundant. SAP explicitly describes using business value, process alignment, and architecture context to identify consolidation opportunities and eventually retire agents that are no longer needed.
So LeanIX is not something that appears at just one point in the lifecycle. It acts more like the architecture layer running underneath it.
And Now Agents Can Use the Architecture Map Too
And after finishing this quick overview of the product, there is another pretty cool functionality we want to highlight.
LeanIX can help companies understand their agents. But, at the same time, AI agents can also use LeanIX to better understand the company and improve their performance.
SAP LeanIX provides an MCP server that allows compatible AI applications to access selected enterprise architecture information. That means an AI assistant can query LeanIX instead of making you manually navigate through architecture documentation every time you need an answer.
Imagine asking:
Which applications depend on this technology?
Or:
Which systems using this component are already scheduled for modernization?
For technical debt work, that can become useful pretty quickly.
An agent could help you identify applications that depend on technologies approaching end of life, then pull the relevant architecture context so you can see which business capabilities could be affected.
Or perhaps you’re preparing to replace an old system. Before changing anything, you could use an AI assistant to explore the applications, interfaces, and other dependencies LeanIX has already mapped.
Basically, the architecture repository becomes much easier to navigate and interrogate.
Of course, that does not mean an agent should autonomously decide to retire an application just because it found a duplicate. Humans still need to evaluate the context and make those decisions.
But it creates a pretty interesting loop:
LeanIX helps agents understand the enterprise, while helping the enterprise understand its agents.
Planning your next SAP project?
Whether you’re mapping your architecture, modernizing your SAP landscape, or adding new AI capabilities, these projects depend on having the right SAP expertise behind them.
As an SAP partner, Inclusion Cloud can provide certified resources to support projects across the SAP ecosystem.
Book a discovery call to meet our specialists and discuss what your project needs.