At SAP Connect on Tuesday, Oct. 6, SAP said Joule Work was beginning its customer rollout, with a select group already using it. SiliconANGLE reported that SAP described general availability for Joule Work and its broader Autonomous Enterprise offering during October. Those statements mark a launch, not proof that the system can reliably handle business workflows at scale.
Businesses should judge the agentic work layer by verified results in real tasks, not by SAP’s Autonomous Enterprise framing. SAP offers reasons to test Joule Work, but the evidence it has presented so far does not establish how often it completes work correctly, how much rework it creates or what customers gain after human review.
SAP’s evidence is promising, but limited
SAP says Joule Work is designed to coordinate tasks across SAP and third-party data and applications. The company says the system draws on its Knowledge Graph, which maps more than 7 million data fields, and describes Joule Assistants as able to act within workflows using existing data, permissions and audit trails. Those are descriptions of the product’s design, not independent findings about its reliability.
SAP also says 110,000 employees use Joule Work internally and reports 20% productivity gains across finance, human resources and procurement. That is a concrete claim, but not yet a decision-grade measure: the figure does not establish which tasks were counted, what baseline was used, or whether corrections and human review were included. The supplied evidence contains no independent evaluation of the result.
SAP’s announcement lists customer examples at different stages. It says Novartis recently launched a pilot of SAP’s Joule Sourcing Assistant, Maschinenfabrik Reinhausen is participating in a Joule Work desktop early-adopter program, and CONA Services is continuing to test Joule. A Reinhausen representative told SAP that a task that previously took up to an hour could be accessed in seconds. That account is worth investigating, but it is not an independent measurement of time saved or errors.
The strongest case for SAP still needs a test
SAP’s best argument is that business context matters. A system connected to company data, workflows and access controls may be more useful than a general-purpose assistant placed over enterprise information. SAP president and chief product officer of Autonomous Suite Manoj Swaminathan told SiliconANGLE, “First principle is it’s AI on apps as opposed to AI in apps”. That distinction explains the product proposition; it does not show that Joule Work performs better than alternatives.
Free newsletter
Get the morning briefing
Start each day with the AI stories that matter and why — a short, free email from our newsroom.
Swaminathan has also described autonomy as a journey in which customers observe and govern agents before delegating work. That is a more credible standard than treating the Autonomous Enterprise name as proof that work can be handed over. But permissions and audit trails alone do not show whether an agent follows the user’s intent, retrieves the right information or knows when to stop and ask for help.
The distinction matters to employees and customers. If Joule Work cuts time spent moving between applications or assembling information, employees may have more time for other work. If it makes an incorrect change, the people using the system and those affected by the business decision bear the consequences. The available examples do not measure those trade-offs across routine use or failure cases.
Matt Cooke, Proofpoint’s cybersecurity strategist for EMEA, argues that organizations should govern agent behavior in context and reserve human approval for sensitive or high-impact actions. His perspective, published in TechRadar, points toward a practical standard: oversight should match the consequences of an action, rather than disappear behind a claim of autonomy.
Require workflow-level evidence before expanding
Businesses should treat Joule Work as a system to evaluate, not a promise to buy into. Before expanding a pilot, CIOs and procurement teams should require results for named workflows: how often tasks finish correctly, how often they need correction, how much time they save after review, and how the system handles uncertainty. The comparison should use a clear baseline and account for the people who check or repair its work.
SAP should make those measures available to customers, and customers should retain approval for actions with significant consequences until tests support broader delegation. They should also require evidence that permissions and audit trails work when an agent acts outside the user’s intent, and establish how to reverse or address consequential errors.
The Autonomous Enterprise may be a useful direction for software design. But the standard for adoption is simpler: show that an agent completes a real workflow reliably, with accountable oversight and measurable value for the people doing the work. A label cannot do that job.