SAP takes Joule outside its own apps with a general release this month

A finance lead gets asked for total corporate spend. In Manoj Swaminathan's own example, the answer is spread across the ERP, the travel and expense system and the procurement platform ^[6]. Joule is supposed to work out which of those it needs and combine the records into one business answer ^[6]. "First principle is it's AI on apps as opposed to AI in apps," Swaminathan, SAP's president and chief product officer of Autonomous Suite, told SiliconANGLE ^[7][15].

SAP is pitching the Autonomous Enterprise. In that version a user declares intent and the agent decides which systems to involve and carries out the task, while a human stays in the loop and agents are treated like employees ^[14]. What ships this month is smaller and easier to plan around: Joule Desktop, working across SAP and non-SAP applications through Joule Work ^[1][4][5]. Swaminathan was direct about how far apart the two are. "When we talk about autonomous enterprise, we realize that it's going to be a journey for the customer, and it's not something you draw a line to say starting tomorrow will be autonomous just because you now have these agents and assistants," he said ^[9].

In SAP's own definition of autonomy, users watch first. According to Swaminathan, people get tools to observe, describe and trust agents before delegating processes to them, along with the transparency to see and audit what the agents did ^[10].

Most of the planning work is in the non-SAP reach. A blended spend answer needs read access to expense and procurement systems whose owners may not sit on the SAP project team. SAP's case for owning that layer is its data model. Swaminathan said SAP organizes customer data into a structured knowledge graph, and that general-purpose models from OpenAI and Anthropic cannot supply ERP semantics, policies and relationships when pasted on top of enterprise data ^[12]. SiliconANGLE's report does not include pricing, licensing terms or customer usage figures ^[3].

For Monday, sort each candidate task on two questions: whether it crosses more than one system, and whether the agent only answers or also acts. Answers from one system are what in-app assistants already handle, and a desktop layer adds little there. Actions inside one system can stay under the approval controls that system already has. The spend question sits in the cross-system answer box. SAP's pitch is strongest there, and the worst outcome is a wrong number someone can check against the source. The last box, agents acting across systems, is where the observe-and-trust step comes due.

I'd start in the cross-system answer box and stay there until the audit trail shows what the agent actually touched. That choice has a cost. The domain agents in this release are the part with the bigger payoff, and under this plan they wait ^[11]. The first quarter's result will be faster reports. Measure the time from question to a trusted answer, and how often users still reopen the source system to check it.