The SAP Approach to Agents
In this blog post series, Niklas Frühauf, Senior Data Scientist, explores whether AI agents already deliver real value for businesses. In part one, he outlined the concept and different types of agents. In part two, he now takes a closer look at SAP’s approach to agents.
SAP has recognized the need to support different types of AI agents, especially given the vast amount of business data stored in SAP systems and their widespread use across industries. At the center of SAP’s strategy is their Joule digital assistant, designed to provide natural-language access to SAP solutions and orchestrate tasks across multiple business processes.
Ready-to-use SAP Agents
SAP continues to expand Joule’s capabilities through ready-to-use pre-built agents for its cloud solution portfolio, which can be explored via the SAP Discovery Center. These pre-built agents natively extend core Joule and SAP solution functionalities without the need for custom code, while SAP ensures that prompt engineering and agent knowledge remain updated alongside the solutions. However, customers on non-cloud or private deployments may have limited access to these agents, often requiring workarounds or custom development.
Custom Agents & Skills for Joule with Joule Studio
With Joule Studio for SAP Build, SAP provides a low-code platform for citizen developers to configure custom agents and create new Joule skills. This allows integration with cloud services, on-premise systems via Cloud Connector and the Destination service, and supports basic AI agent governance including versioning, release management, and access controls for different users. Here at sovanta, we like using this approach for customers that are already working with Joule and Build. For Piller Blowers & Compressors GmbH, we built a spare parts agent that autonomously creates budgetary offers based on the information provided by the customer, while asking follow-up questions whenever additional details are required. (Project submitted for SAP Innovation Award) Together with Endress + Hauser, we are rethinking the customer engagement process: a sales assistant agent that can access customer data, service incidents, past orders, and the company’s data warehouse helps sales teams prepare for customer calls and enables them to run ad-hoc follow-up questions to identify purchasing trends, anomalies, and patterns.
One notable limitation of Joule’s bundled agents and skills is interoperability. While Joule can easily call third-party agents or MCP tools, its capabilities can currently only be invoked from within SAP solutions by users interacting directly with Joule. Plans exist to extend this to event-based triggers or API calls, but these features are not yet available.
Ultimate Flexibility: Pro-Code Agents on BTP
Finally, for tasks requiring more advanced or specialized reasoning, SAP supports pro-code agents. Developers can build these agents in their preferred language (e.g., using LangGraph/Chain with CAP/TypeScript apps or LlamaIndex/LangGraph in Python) on Cloud Foundry or Kyma runtime. These retain full flexibility and can be accessed externally, enabling integration with non-SAP systems or custom workflows beyond Joule’s immediate interface.
These pro-code agents can be exposed as additional Joule capabilities, enabling Joule to delegate tasks to them as needed, effectively combining low-code orchestration with fully custom agent logic. In the retail and grocery sector, we developed a pro-code agent that processes incoming emails and automatically determines which teams or employees should handle them, taking into account employee availability, staffing, skills and vacations. The agent runs on a scheduled basis, highlighting another advantage that pro-code agents can bring: We are not depending on external scheduling services or triggers. We also built another agent that collects and enriches product master data from various unstructured sources and automatically updates it in the relevant target systems. Last but not least, we added an agentic assistant to one of our partners’ integration platforms. The assistant helps engineers quickly perform root-cause analysis for failed integration scenarios by reviewing configuration settings, analyzing optional custom code written by developers, verifying system availability, and correlating errors across different integration flows to suggest potential solutions.
SAP Roadmap: What’s to Come?
- SAP is continuing to expand the portfolio of domain-specific agents within the SAP Joule ecosystem. Based on SAP’s extensive experience with industry-specific business processes, planned agents cover a wide range of business functions, including financial accrual proposals, order reliability analysis, bid and tender evaluation, expense automation, payroll alert resolution, and HR analytics. These capabilities are being embedded directly into SAP business applications such as SAP S/4HANA Cloud, SAP Ariba, SAP SuccessFactors, and SAP Concur. The goal is to provide ready-to-use AI assistance for common business workflows while maintaining tight integration with application data and processes.
- For citizen developers, the roadmap expands the ability to build and extend Joule capabilities through SAP Build Joule Studio and SAP Build. Planned improvements include standard agent extensibility, allowing organizations to adapt SAP-delivered agents with custom tools and business logic while maintaining upgrade compatibility. SAP is also introducing deeper integration between Joule agents and workflow automation, enabling process automation scenarios to trigger agents directly. Additional capabilities—such as centralized management of external tools via MCP hubs and event-based triggering of agents—aim to support more autonomous automation patterns. In parallel, SAP plans tighter governance integration with SAP LeanIX, allowing organizations to discover and manage AI agents and extensions as part of their broader enterprise architecture landscape.
- For pro-code developers and data scientists building custom agents on SAP BTP, SAP is strengthening the underlying AI infrastructure centered around SAP AI Core and the SAP Generative AI Hub. Planned improvements include enhanced grounding capabilities that incorporate external web sources as well as dynamic model routing that automatically selects the most appropriate LLM for a given task. SAP is also expanding support for agentic AI scenarios within SAP S/4HANA through intelligent scenario lifecycle management for ABAP developers. To support production deployments, the roadmap includes improved observability and monitoring—for example through agent observability features in SAP Build and deeper integration with SAP Cloud ALM as well as mechanisms such as Model Context Protocol (MCP) endpoints that allow agents to securely consume enterprise APIs and services managed via SAP Integration Suite.
There clearly is a lot of drive behind the different agent topics at SAP – and so is customer demand.
What’s next?
But what is holding companies back from going all in on agents? Why do we still see so few agents in action? That’s exactly what the third and final part of this blog post series explores.
