SAP has used its TechEd developer conference to widen what its Joule agents can do and to put rules around how they are reached. The announcements, made on 27 October, cover Joule Work, Joule Studio and the SAP Business AI Platform, and they fit the company's Autonomous Enterprise push.
Joule Work gains support for MCP and Agent2Agent, so agents can collect context, make decisions and run multistep workflows across systems, including third-party data and services. A new orchestration layer lets other assistants and agents call Joule Agents through a governed gateway, where their activity can be enforced and overseen. Joule Studio agents can also trigger n8n workflows, which for now is open to early-adopter customers, with broader availability to follow.
SAP's Ted Way explains the thinking: agents should be governed centrally, much as employees are, with access and tool permissions, and customers' own agents should reach SAP's through a controlled entry point. He says n8n suits regulated work that needs exact steps in a set order, so SAP pairs flexible natural-language triggers with prescribed workflows.
“Open protocols need governed doors”
Ted Way, VP and chief product officer, SAP business AI product engineering, to SiliconANGLE · siliconangle.com ↗
The timeline matters for reading these announcements. SAP introduced the Autonomous Enterprise at its Sapphire conference in May 2026, and Joule Work reached all customers from October, after SAP Connect. TechEd is the developer-facing follow-up to those events, which is why so much of it concerns protocols, runtimes and the tooling that builders use rather than finished applications.
Joule Studio offers more than 70 AI models and no-code and pro-code tools, and gains observability and admin controls showing agent behaviour, model usage and spend. Running custom agents on SAP's managed runtime stays free through December 2026 for selected customers, and design-time tooling stays free under fair use.
The free runtime offer has an end date. Custom agents on SAP's managed Joule Studio runtime cost nothing for selected customers only through December 2026, and broader n8n availability in Joule Studio is promised without a date. Customers outside the early-adopter group will have to wait for it, and SAP has not said how long.
The data side gets work too. Business Data Cloud's Ontology Manager adds AI-native authoring of ontologies across SAP and third-party data. The Dremio-based lakehouse gains native Apache Iceberg and Apache Polaris support and a federated query engine that avoids copying data, and SAP Reltio, from the acquisition announced around SAP Connect, brings multidomain master data management. The Databricks partnership deepens, bringing Lakeflow, Agent Bricks, Unity Gateway and Lakebase into Business Data Cloud, with a Joule and Genie integration.
Stock image heading SAP's piece on running its tabular AI model on the International Space Station
Getty Images (per file name; page prints no credit) · Image: SAP News Center
The most unusual item is a tabular model. SAP-RPT-2.0 adds anomaly and outlier detection and embeddings, and a new SAP-RPT-TS forecasts time-indexed metrics through Integrated Business Planning without custom machine learning. SAP says its SAP-RPT family is about 50,000 times more energy-efficient than leading LLMs. With HPE it plans to run SAP-RPT on the Spaceborne Computer-2 aboard the International Space Station by early 2027, for anomaly triage, failure prediction, rebuilding missing sensor values and mission-health reports. Way's reasoning is that much business data sits in tables, so language models alone are not enough.
SAP's framing of the tabular work is that a language model is the wrong tool for much enterprise data. Way's line is that not every enterprise problem is a language problem, and the company pitches its tabular models as a lighter alternative for forecasting and anomaly detection on rows and columns. The space deployment is presented as a stress test: a spacecraft's sensors produce exactly the kind of time-indexed numbers the models are meant for, and the plan is for them to flag problems, predict failures and fill in sensor values that go missing, with mission-health reports on top.
The space plan is still a plan: deployment is set for early 2027, and the energy comparison with leading LLMs is SAP's own claim, made in SAP's own announcement and not a figure from an outside test.
About Joule
Joule is SAP's AI assistant and agent layer for its business software. Joule Work is the single experience where users state an intent and get work done across the SAP landscape, and Joule Studio is the AI-first environment for building and extending enterprise agents.
SAP presented its Autonomous Enterprise at Sapphire in May 2026 as the next step for Joule, describing a rebuild from a natural-language chat client into a new engagement layer. That month it made Joule Work available to Early Adopter Care customers. By its October Connect event it said it had built more than 20 Joule Assistants and 200 Joule Agents since Sapphire, was aiming for 400 agents by year end, and was rolling Joule Work out to all customers.
Partner extensions were announced with Docusign, Parloa and Redwood, and an Intel collaboration links workflows to autonomous robots. SAP also says customer data stays with customers and is not shared to train LLMs. SAP's CTO Philipp Herzig frames the shift as AI moving from learning a business to running it, with governance built in from the first deployment.
Illustration of two business-dressed AI agents talking to a prospective client
Image: SiliconANGLE