TL;DR — Key Takeaways
– SAP is rolling out Joule Work, a framework designed to let AI agents and applications work across SAP and third-party data to automate enterprise workflows.
– SAP says its catalog of AI agents will grow to more than 400 by the end of 2026, with Joule Work helping assign tasks to the appropriate agents through natural language.
– The company is using a knowledge graph to constrain the data available to AI agents, with the goal of improving the accuracy and reliability of agentic workflows.
SAP today revealed it has begun to roll out Joule Work, a framework that enables artificial intelligence (AI) agents and applications to invoke both SAP and third-party data, to provide organizations with a foundation for building an autonomous enterprise.
Additionally, SAP announced it has acquired TechWolf, a provider of an AI platform that tracks, analyzes, and updates employee skills and the tasks that have been assigned to them.
Finally, SAP also disclosed it is extending its reach into payments with the launch of SAP Pay, an offering based on a platform from Tereina that SAP holds an equity stake in.
Speaking at the SAP Connect 2026 conference, SAP CEO Christian Klein told conference attendees that Joule Work is a user interface for AI that has been trained to understand how SAP applications work. At the core of that new engagement layer for SAP applications is a knowledge graph the company developed to ensure agentic AI workflows based on SAP data are accurate, he added.
Armed with that capability, it then becomes possible via a natural language interface to assign tasks to SAP agents that will exceed 400 by the end of this year, says Klein.
“AI can be impressive but it is not enough,” he says. “It needs to be accurate.”
There are currently more than 400 million users of various SAP applications but it’s not clear what percentage of them will be relying on AI agents developed by SAP to automate workflows. While SAP has limited access to its APIs, many organizations in the AI era are deploying AI agents they either licensed elsewhere or built themselves to drive agentic AI workflows.
SAP, alternatively, is making a case for a set of AI agents that can be orchestrated and managed via a Joule Work framework that automatically assigns tasks to them based on the process that needs to be completed.
The issue that SAP is fundamentally trying to address is the simple fact that most business workflows are deterministic in the sense they are expected to be completed the same way every time. AI agents and applications, in contrast, are probabilistic by definition, which means they almost never perform the same task the same way twice. By using a knowledge graph limiting the scope of the data that AI agents can access, SAP is making a case for an approach to generative AI that will result in more reliably accurate outcomes. Organizations that have already adopted SAP AI agents include Novartis, Morgan Foods and Nestlé.
Just how fully autonomous any large enterprise is likely to become remains to be seen. For the most part, organizations have been able to leverage AI agents to create multiple instances of semi-autonomous workflows that still require humans to be somewhere in the loop. One way or another, it’s not so much a question of whether AI agents will be incorporated into business workflows as much as it is the degree to which they will be trusted to automate a task. As such, the one thing that is certain is there will undoubtedly be a lot of trial and error as organizations test the limits of not only what AI agents can do today but also tomorrow as advances in the development of large language models (LLMs) continue to push the art of the possible.
