At SAP Connect in Las Vegas, SAP SE (NYSE: SAP) explained to its customers and potential customers why they should choose SAP as their partner for Agentic AI to drive the autonomous enterprise. SAP is the largest global provider of enterprise applications.

SAP’s agentic framework is known as Joule: “Joule Assistants” help streamline work, and the larger agentic framework is called “Joule Work.” Assistants can be asked why something happened, used to analyze and suggest solutions, and, in the future, customers may even trust them to execute tasks. Novartis, for example, co-developed and recently launched a pilot of SAP’s Joule Sourcing Assistant. Their vision is for AI agents to handle the manual heavy lifting, enabling sourcing professionals to focus on strategy and supplier engagement.

Wall Street is looking for progress on the AI front from SAP. In August, UBS downgraded the enterprise software giant from Buy to Neutral, citing a glaring gap between the company’s grand AI ambitions and its slow rollout of actual deliverables. While SAP may have over promised, they are not behind other supply chain software companies I cover.

Since SAP Sapphire in May, SAP has developed 20 assistants. These assistants are role-driven and deeply embedded in a process workflow. They coordinate multiple agents to do their work. Assistants span procurement, supply chain, finance, customer experience, human resources, and Industry AI.

Industry AI is necessary because different industries can have very different processes and measures of success. A logistics service provider works very differently than a chemical manufacturer.

Customers on stage did not talk about the ROI they have achieved; most are still working to co-develop agents with SAP or, in Novartis’s case, running a pilot. ROI also depends upon what the agents cost. The cost will be based on a usage model, SAP’s version of token pricing. Klein did note that they support a wide variety of LLM solutions. Paying for the most advanced frontier model will often not make sense.

One beta customer I talked to complained about the slow development of their supply chain agents. SAP is co-developing the agents with them, which they can try out for free in the short run. In the longer run, SAP will charge for the usage, but those charges are not yet clear.

The Knowledge Graph is Key to Business Value

SAP is right that the knowledge graph is a clear differentiator against roll-your-own companies like Palantir , which use “forward-deployed engineers” to build a context layer for large language models before rolling out a custom agentic AI solution for customers.

Christian Klein, SAP’s CEO, put it this way, “Frontier models have gotten better, but here’s the thing: ask the model to draw a unicorn, and you get a great picture. But if you ask the model which of your suppliers will deliver next week, it will guess. LLMs still have no idea how your business runs.” In contrast, “Our knowledge graph maps more than 7 million data fields, half a million tables, over 50,000 APIs, as well as our private APIs, 400 data products, and our SAP process knowledge.”

Klein said later, “But how can our AI assistant guarantee the right answer when LLMs are probabilistic and only guess the most likely answer? Now, because our AI agents are embedded in the deterministic workflows of our applications, they deliver the one right answer because they are running on the same data, (and) they follow the same rules.”

I’m not sure SAP agents will always deliver the right answer. Assuming a customer’s data is clean, the assistants do go to the right location on the graph and pull the right data. But part of training these models is ensuring they fully understand the question that is being asked. I’ve noticed that with LLM search engines, how a question is phrased determines whether I get the right answer; the phrasing often determines where the answer comes from.

SAP developed its knowledge graph for its newer Cloud solutions. Older, on-premises versions don’t have an out-of-the-box knowledge graph. Building these graphs for every older release would be too big an investment.

SAP will extend its agentic framework to these older versions, or even other enterprise software solutions an enterprise might also be using; this is where it employs its own version of “forward deployed engineers” to build a context layer that supports these legacy solutions. When SAP talks about “extensibility,” this is what they mean.

For companies eager to use the newest generation of enterprise solutions, agentic solutions grounded in how their business actually works could strongly motivate a move to the new Cloud solutions. Once customer ROI data is available, Joule could drive a new wave of upgrades.

AI Governance is a Differentiator

While the knowledge graph differentiates SAP from “roll-your-own” AI system integrators, their approach to governance is their main differentiator against other enterprise software suppliers.

SAP runs and governs every agent on the SAP AI platform. SAP ensures identity and authentication management for every AI agent. Just as software systems can be hacked, a cybercriminal could possibly gain control of an agent. Imagine a transportation agent, for example, that is told to deliver a shipment of valuable goods to a warehouse operated by thieves. That is the value of identity and authentication management.

But SAP agents are also being built to comply with regulations in over 60 countries and 26 industries. They’re built to be certified by a firm’s auditors. This is, of course, valuable for financial agents that need to follow banking rules. But in the supply chain realm, cross-border shipments must follow the tariff rules of many different nations; other assistants must follow different countries’ product quality and recall rules; and customer assistants must follow global rules protecting customer data.

AI Governance is the area where SAP leads most of its enterprise software competitors.