The news
SAP (SAP) is turning its Joule assistant into what it calls an agentic work layer across its business software. At its SAP Connect 2026 event on October 6, the company said its Autonomous Enterprise architecture, first introduced at its Sapphire conference in May 2026, becomes generally available in October, SiliconANGLE reported. SAP's own release does not state a general availability date and stresses rollout to early adopters.
The announcements center on Joule Work, a new AI user interface that lets employees ask for work in their own language and complete tasks across SAP and third-party data, according to SAP's press release. SiliconANGLE also reported a Joule Desktop offering, and SAP's release refers to a Joule Work Desktop early adopter program. SAP said Joule Work is in customer rollout with select early adopters.
SAP also introduced Joule Assistants, which carry out operational work inside business workflows using existing data, permissions and audit trails, and updated agents for finance, supply chain, spend, and workforce and customer experience. SAP said its International Trade Assistant can cut trade classification effort by up to 50%.
SAP said 110,000 of its own employees use Joule Work and that it has seen 20% productivity gains across finance, HR and procurement. Early customers named by SAP include Novartis, Morgan Foods, Nestle, PwC, Maschinenfabrik Reinhausen and CONA Services. SAP chief executive Christian Klein said in the release that the Autonomous Enterprise has arrived.
Manoj Swaminathan, president and chief product officer of SAP's Autonomous Suite, told SiliconANGLE the approach is AI on apps rather than AI inside each app, and that Joule is no longer just a chat client but a full-screen experience. SiliconANGLE reported that SAP charges for this mainly through AI units, a consumption-based model in which agent inference uses up units, and that customers can see their usage.
The numbers
- SAP employees using Joule Work
- 110,000
- Productivity gain SAP reports internally (finance, HR, procurement)
- 20%
- Cut in trade classification effort, per SAP
- Up to 50%
- Data fields mapped in SAP Knowledge Graph
- 7+ million
Why CEOs should care
For CIOs running SAP, the question is no longer whether agents will touch ERP, but which processes you will let them run and under what controls. SAP stresses that its assistants work within existing permissions and audit trails. Ask your SAP account team which agents are generally available versus in early adopter rollout, which approvals stay with humans, and how agent actions show up in audit logs your external auditors already rely on.
For CFOs, the pricing model is the headline. Consumption-based AI units mean cost scales with use, which can be hard to forecast once agents run continuously. Ask for unit pricing, how many units typical tasks consume, whether there are caps or alerts, and how units are bundled in RISE contracts. Treat SAP's 20% internal productivity figure as a vendor measurement on its own staff, and run your own pilot with a baseline before budgeting savings.
For boards, letting software act inside finance and supply chain systems is a control question. Ask management who owns agent errors and how quickly an agent can be switched off.
The bigger picture
Enterprise software vendors are racing to make their own assistants the front door to work, so customers do not route around them through general chatbots. SiliconANGLE framed general-purpose models from OpenAI and Anthropic as the competition, arguing that SAP's knowledge graph of business context gives Joule an edge.
The shift to consumption pricing mirrors moves elsewhere in software, where per-seat licenses fit poorly with agents that do work without a human user.
What’s next
Watch for customer reports on AI unit consumption after general availability, the pace at which early adopter features such as Joule Work reach all customers, and whether SAP publishes clearer price lists for agent use ahead of 2027 budget cycles.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
