Need to automate invoice exceptions? Build an agent.
Need to analyze supplier risk? Build another one.
Need to help planners react to a delayed shipment? Another agent.
Need something for HR, finance, customer service, or procurement? There will probably be an agent for that too.
This is where enterprise AI is heading. SAP itself is preparing an Autonomous Suite with more than 50 Joule Assistants and more than 200 specialized agents across finance, procurement, supply chain, HR, and customer engagement. The assistants can orchestrate subsets of those agents to execute larger business processes.
And those are only SAP’s agents.
Your company may also use agents from Microsoft, Google, ServiceNow, Salesforce, AWS, or internal development teams. Some may use SAP models, others Claude or another LLM. Some may interact with SAP data, while others connect through APIs or MCP servers.
By building SAP AI Agent Hub as a vendor-agnostic product, SAP is acknowledging an important reality: the enterprise AI stack is likely to be multi-agent and multi-model.
That creates another problem.
How do you keep track of everything? That is the problem SAP AI Agent Hub is designed to address.
What Is SAP AI Agent Hub?
SAP describes AI Agent Hub as a vendor-agnostic command center for discovering and governing AI agents, LLMs, and MCP servers across the enterprise. It creates a structured inventory of these AI assets and connects them with the business and technology landscape around them.
There is a little history behind the product.
An earlier version of AI Agent Hub was already generally available through SAP LeanIX. At Sapphire 2026, SAP announced a much larger revamp as part of the new SAP Business AI Platform. CIO reported in May that the existing Hub was already available while SAP was working on this expanded version. SAP, meanwhile, announced that the new governance layer would roll out during Q3 2026.
So we are currently in the middle of that transition.
The important change is scope. SAP wants Agent Hub to become a central place for governing not only SAP agents, but the broader AI landscape around them.
Why Do You Need an Agent Inventory?
Agent governance starts with a surprisingly basic question:
What agents do we actually have?
But answering that question could be harder than it sounds…
Finance may deploy one agent. Procurement builds another. Marketing experiments with a third-party platform. A developer creates a custom agent for an internal team. Six months later, another department builds something very similar without knowing the first agent exists.
SAP and Oxford Economics found that only 44% of businesses currently maintain an agent registry, while only 32% have cost or usage controls for agents.
SAP AI Agent Hub tries to create that inventory automatically.
Discovered agents enter a central repository where you can connect them with owners, business capabilities, applications, models, and other relevant information. SAP also provides verification workflows so organizations can distinguish agents that have gone through their governance process from those that have not.
This matters from a tokenomics perspective, but security is the bigger concern.
An Agent Can Only Be as Safe as Its Access
Imagine you build an HR agent to help managers plan staffing.
The agent needs access to:
- Employee roles,
- Schedules,
- Availability,
- Organizational structure.
But somebody gives its integration account broader permissions than necessary.
The agent can now access payroll and individual compensation records too.
A manager asks:
“Why is my team’s staffing cost so high this month?”
The agent finds the salary data and includes individual compensation in its answer.
The model did not necessarily “hack” anything. It accessed information that its credentials allowed it to access.
The governance problem was that the agent never needed those permissions in the first place.
The SAP/Oxford Economics study shows how incomplete those controls remain. Only 63% of surveyed organizations have permissions and access controls for agents, 53% have approval processes before deployment, and half maintain audit logs of agent decisions.
In a CIO.com interview, Sean Kask, SAP’s Chief AI Strategy Officer, describes agent governance as a “hire-to-retire” lifecycle: discover and register agents, assign permissions and data access, observe their behavior, evaluate their performance, and eventually retire them.
That is a useful way to understand what SAP is trying to build with Agent Hub.
What Can You Govern in Agent Hub?
SAP’s current documentation organizes the Hub around three simple questions:
1) What agents do we have?
Automated discovery brings agents from different platforms into one inventory.
2) Are they governed?
You can assign ownership, classify risk, verify agents against governance criteria, and track their lifecycle.
3) Are they delivering value?
SAP can associate agents with expected cost savings, revenue impact, risk reduction, usage, and other business outcomes.
The Hub also goes beyond agents themselves.
It can represent the AI models powering those agents and the MCP servers they use to access external tools and data.
In practice, you are governing something closer to:
Agent = purpose + model + instructions + data + tools + permissions + owner + execution
If one of those pieces changes, the agent’s risk or cost can change too.
How Does Agent Hub Work With Other SAP Tools?
Agent Hub is not supposed to do everything itself.
SAP is connecting it with other parts of its stack, and each one adds a different layer.
SAP LeanIX: Architecture Context
Agent Hub is built on the SAP LeanIX data model.
If you already use SAP LeanIX Application Portfolio Management, Agent Hub is available directly within the workspace. LeanIX adds the broader enterprise architecture view around each agent: applications, business capabilities, technologies, organizations, and dependencies.
That helps you move beyond:
“We have this procurement agent.”
toward:
“This procurement agent supports this business capability, depends on these applications, uses this model, and interacts with these systems.”
Joule Studio: Build
Think of Joule Studio as one of the places where you create agents and Agent Hub as the place where you govern the resulting portfolio.
SAP is also opening Joule Studio to code-based agents and third-party frameworks, which makes centralized governance more relevant as companies gain more ways to build agents.
SAP Cloud ALM: Runtime Observability
Inventory tells you that an agent exists.
Observability tells you what it actually did.
SAP is connecting Agent Hub with SAP Cloud ALM so teams can inspect execution behavior, including the steps, tools, systems, and data involved in an agent session. SAP has said it is expanding runtime observability as part of the Agent Hub roadmap.
That becomes useful when an agent produces a bad result or suddenly becomes inefficient.
SAP Signavio: Process Context
SAP is also connecting agent governance with process intelligence.
Suppose an invoice exception agent works correctly but takes nine steps to resolve something that another execution path handles in four.
Looking at the process can help you identify that inefficiency.
This is where Agent Hub starts crossing into AI FinOps: sometimes reducing AI cost is not about finding a cheaper model. It is about fixing an inefficient workflow.
Identity and Access
SAP is also expanding identity and access controls around Agent Hub. The objective is to connect agents with enterprise identity and authorization rules rather than treating them as anonymous pieces of software.
That matters for the HR example above.
The question is not simply whether the agent technically can read payroll.
It is whether it should.
Where Does Tokenomics Enter the Picture?
This is the part I find most interesting.
AI Agent Hub is not an AI Unit billing dashboard.
If your question is:
How many AI Units did we consume this month?
you still need SAP’s consumption and commercial management tooling.
Agent Hub answers a different set of questions:
- What agents are behind that consumption?
- Who owns them?
- Which models do they use?
- How often are they being used?
- How efficiently do they work?
- And are they generating enough value to justify their cost?
Kask makes this connection explicitly. He argues that companies should monitor whether an agent achieves its intended result, how efficiently it works, how many tokens it consumes, and how many processing steps it requires for a task.
That is AI FinOps at the agent level.
Consider a procurement agent that normally completes a task in four steps.
After several changes, the same task now takes nine:
- Retrieve supplier information,
- Call an LLM,
- Search another system,
- Call the model again,
- Retrieve a document,
- Summarize it,
- Call another tool,
- Make another model call,
- Generate the recommendation.
The final answer may still be correct. But your execution path has become more expensive.
If you can see that behavior, you can start asking useful questions:
- Can we remove some steps?
- Can we use a smaller model for part of the workflow?
- Can a deterministic process replace one of the reasoning steps?
- Does this workflow even need an agent?
Agent Sprawl Is Also a Cost Problem
There is another form of waste that token dashboards alone will not show you very clearly: duplication.
Imagine Sales builds a customer research agent.
Marketing builds another.
Customer Success builds a third.
They use slightly different prompts and models, but all three retrieve similar information and solve essentially the same problem.
Each one generates its own consumption.
The problem is no longer simply:
“Which model is cheapest?”
It is:
Why are we paying for three agents?
SAP’s own research suggests this is already happening. Only 31% of respondents disagreed that different teams were building overlapping agents for the same needs.
A central inventory gives you a chance to spot those overlaps before they become permanent parts of your AI stack.
And sometimes the right FinOps decision will be simple:
Decommission the agent.
Cost Is Only Half of the Equation
Reducing consumption is useful, but the cheapest agent is not necessarily the best one.
Imagine:
Agent A costs $20,000 per year and creates $500,000 in measurable savings.
Agent B costs $5,000 and nobody really uses it.
Agent A costs four times more.
It is also clearly the better investment.
This is why AI FinOps eventually has to move from cost visibility to value visibility.
SAP is already moving in that direction. Agent Hub’s current documentation includes business-value calculations tied to expected cost savings, revenue increases, and risk reduction.
This also connects with a broader problem companies have had with technology for decades: measuring business value is difficult.
SAP and Oxford Economics found that companies are spending more on AI while still asking what value they are actually getting from it. The same report argues that organizations now face two challenges at once: governance risks are moving quickly while AI value remains difficult to measure.
Agentic AI makes both problems harder because one business outcome can involve several agents, models, tools, systems, and human approvals.
Knowing the cost of every model call does not automatically tell you which part created the value.
But having a structured inventory gives you somewhere to start connecting the two.
How Do You Access SAP AI Agent Hub?
SAP currently documents two access paths.
The first is through SAP LeanIX Application Portfolio Management. Agent Hub is included by default in those workspaces, with no additional provisioning required.
The second is SAP AI Agent Hub standalone, focused specifically on AI governance without the full enterprise architecture management capabilities of LeanIX.
SAP Help says standalone eligibility applies to SAP Business AI Platform customers receiving Joule Studio or AI Units. The current standalone access documentation still references the Early Adopter Care program, whose registration is now closed, reflecting the fact that SAP is still moving through the Q3 rollout of the expanded product.
So if you are evaluating it right now, availability can depend on your existing SAP products and rollout status.
The Bigger Picture
As enterprise stacks become more autonomous, the gap around governance and cost control gets wider.
More agents mean more models, more tools, more data access, more permissions, and more execution paths to keep track of. At the same time, AI costs can spread across teams and use cases quickly, especially when different departments build overlapping agents or use different platforms.
That creates two problems at once.
You need to understand what each agent is allowed to do.
And you need to understand what each agent is costing you and whether that cost is justified.
And SAP AI Agent Hub is one piece of that puzzle.
If your SAP landscape is moving in that direction, Inclusion Cloud is an SAP partner that can help you strengthen governance and optimize AI costs, with certified SAP consultants supporting development, integrations, migrations, deployment, support, and the scaling of SAP environments. We also support organizations across regulated industries and public services, where access controls, traceability, compliance, and cost visibility become even more important as AI agents take on more business processes.
Book a call with our team to review your SAP AI landscape and identify where you need to refine governance or optimize costs.
