SAP is adding detail to its vision of an “Autonomous Enterprise,” describing how businesses could manage the costs, access and oversight of AI agents working within SAP systems. At the TechEd developer conference in Berlin, the company outlined capabilities associated with Joule Work and SAP Autonomous Suite, built on the SAP Business AI Platform. The emphasis is not only on AI answering questions, but on agents that can analyse information and carry out tasks in business software—making the rules governing their use a central part of the plan.
Budgets tied to expected use
SAP Chief AI Officer Jonathan von Rüden said customers first need to decide which problems they want AI to address. He described three broad areas: everyday productivity tasks such as summaries and research; centrally managed agents, including SAP’s standard options and customer-built agents; and specialised applications that could help a company distinguish its operations. The distinction matters because broad employee use is harder to forecast than a clearly defined, repeatable process.
To estimate consumption, SAP considers the number of employees, how often a process occurs and the volume of documents involved. It also draws on experience with comparable customers and industries, von Rüden said. Those estimates provide a basis for budgets, but actual costs still depend substantially on use. SAP’s controls are intended to put boundaries around that variable demand rather than assume that a single forecast will accurately predict every user’s activity.
In Joule Work, businesses can set a daily usage allowance for each employee, with an indicator showing how much has been used. Budgets can also differ by department or country. When someone reaches a limit, they may request more allocation if a task needs to be completed. That approach combines usage-based billing with central limits while leaving a route for work that exceeds its initial allowance. SAP also says customers can optionally analyse agents and processes to check whether the intended benefits have materialised.
Model choice and routing
Model selection is another cost lever. Von Rüden said users often keep the default model, so choosing a default that balances capability and cost can affect consumption without requiring employees to make a fresh technical decision each time. Budgets may also prompt users to consider whether a more powerful, and potentially more expensive, model is necessary for a particular task. SAP supports 70 large language models in Joule Work, according to the interview.
For users, model options are presented through modes such as Standard, Extended and Light, rather than requiring everyone to select a specific model version. Developers building their own agents can choose from available models. SAP’s teams evaluate new models for existing assistants before deciding whether to replace the models behind them, and the company also considers regional requirements, including customer preferences for European models.
SAP is testing automatic model routing internally and plans to offer it to initial customers in the future; it is not described as generally available. Von Rüden said routing becomes more difficult when an agent must find and connect transactional business objects, compared with a simpler task such as summarising documents. The system must match the model to the work involved, so the company is still evaluating how to make that selection reliably.
Keeping agent development governed
SAP’s controls also address who can create agents. Access to Joule Studio, the tool for building them, is assigned by IT rather than automatically granted to every employee. SAP is initially making the tool available to a limited group of employees outside IT and development, including people with relevant familiarity and training. Companies can set their own access rules, giving them a way to restrict development while they establish internal practices.
Not every personal adjustment has to become an organisation-wide agent. An employee might first tailor how an agent delivers results for their own workflow; if the idea could help others, they can discuss whether to develop and share it or pass it to a Center of Excellence. Joule Studio supports sharing with selected colleagues for testing, while roles and an identity provider can control broader access. SAP’s AI Agent Hub provides an inventory showing agents’ review status, and the lifecycle can include company-specific approvals. Taken together, these controls reflect the operational challenge behind SAP’s vision: giving teams useful automation while keeping costs, access and accountability visible. The company’s comments describe a mix of existing controls and capabilities still being developed, including automatic routing, so businesses will need to distinguish what they can use now from what SAP plans to deliver later.