{"id":240,"date":"2026-11-18T21:13:34","date_gmt":"2026-11-18T21:13:34","guid":{"rendered":"https:\/\/blockoala.com\/index.php\/2026\/11\/18\/sap-executives-ai-agents-need-business-context-and-controls-before-they-can-run-core-workflows\/"},"modified":"2026-11-18T21:13:34","modified_gmt":"2026-11-18T21:13:34","slug":"sap-executives-ai-agents-need-business-context-and-controls-before-they-can-run-core-workflows","status":"publish","type":"post","link":"https:\/\/blockoala.com\/index.php\/2026\/11\/18\/sap-executives-ai-agents-need-business-context-and-controls-before-they-can-run-core-workflows\/","title":{"rendered":"SAP executives: AI agents need business context and controls before they can run core workflows"},"content":{"rendered":"<p>In an interview reported on November 17, 2026, SAP executives described a gap between impressive AI-agent demonstrations and reliable deployment in enterprise operations. A proof of concept may work with prepared data and a defined task, said Cai Yanbin, president of SAP China Research Institute, only for inaccurate data and other problems to emerge in production.<\/p>\n<p>The discussion followed SAP TechEd 2026, where agents\u2019 potential role in core business processes was a prominent topic. SAP\u2019s proposed \u201cAutonomous Enterprise\u201d vision would move AI beyond assistance toward coordinating and executing work. But the executives said existing systems, data and business rules must be made usable by agents first.<\/p>\n<h3>Business context is the hard part<\/h3>\n<p>Company data is often spread across ERP, finance and supply-chain systems. Even a term such as \u201ccustomer\u201d can mean the contracting party in one system and the recipient or payer in another. Cai said an agent cannot decide on its own which system\u2019s rules should prevail.<\/p>\n<p>He argued that deployment requires standardized data and processes, master-data governance and what he called <em>semantic readiness<\/em>: a way for AI to understand what enterprise data means and how it relates. SAP is working to represent business data and processes in a knowledge graph. According to the report, its ERP systems encompass more than 7 million fields, over 500,000 data tables and more than 15,000 business processes.<\/p>\n<p>A budget-analysis example illustrates the intended use. Given a scenario in which fourth-quarter electricity prices are expected to rise 20%, Joule Work draws on company data to identify six cost centers, flags three as potentially facing shortfalls, and suggests checking when the price increase takes effect, departmental electricity use and possible additional funding. That is an example of the system\u2019s approach, not evidence that agents can independently resolve every budget decision.<\/p>\n<h3>Agents alongside ERP, not in place of it<\/h3>\n<p>Cai expects more employees to request work through an AI interface rather than navigate ERP transaction codes and screens. He also said conventional interfaces will remain necessary when people must intervene or handle exceptions. In SAP\u2019s proposed architecture, agents interpret requests and coordinate tasks, while existing applications continue to run finance, HR and supply-chain processes.<\/p>\n<p>Cai said SAP had released more than 300 agents. Joule Work serves as a user-facing task interface, while Joule Studio is intended for building agents, including from natural-language descriptions of business needs. Easier development, however, creates a maintenance problem: an update to one agent could disrupt others that depend on it in a purchasing or approval workflow.<\/p>\n<p>SAP wants agent catalogs, version management and controls for connections using A2A, MCP and APIs. Cai called for testing functionality, performance, security, accuracy and compliance before deployment, followed by ongoing monitoring. He said enterprises should establish those processes before hundreds of agents are running.<\/p>\n<h3>Limits on autonomy<\/h3>\n<p>Clas Neumann, SAP\u2019s global head of research and development network, said organizations must give AI some scope to act while keeping people responsible for setting rules and outcomes. Cai said agents are currently better suited to execution than management or strategy. His suggestion that agents might eventually handle 70% to 80% of execution work in some scenarios was a possibility dependent on improved reliability, not a current result.<\/p>\n<p>Cai said SAP aims to extend existing ERP user permissions to agents, so an agent cannot read or change information beyond the authority of the user it represents. Enterprises also need records of what an agent accessed and changed, along with ways to investigate errors and, where appropriate, reverse actions. The method of reversal would depend on the business process, he noted.<\/p>\n<p>SAP also described a warehouse collaboration with Intel\u2019s Chengdu factory in which business agents pass picking and delivery tasks from a warehouse management system to robots and return results to enterprise systems. SAP\u2019s estimate that the approach could reduce manual work by about 90% is a projection for that solution, not a demonstrated result across warehouses.<\/p>\n<p>Readiness remains limited. SAP\u2019s <em>2026 SAP AI Value Report<\/em> found that only 6% of Chinese companies surveyed said they were fully prepared for agentic AI. Cai said integrating AI throughout core business operations still has a long way to go: companies first need to organize their accumulated data and processes into a form agents can understand and use.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>In an interview reported on November 17, 2026, SAP executives described a gap between impressive AI-agent demonstrations and [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-240","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/blockoala.com\/index.php\/wp-json\/wp\/v2\/posts\/240","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blockoala.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blockoala.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blockoala.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/blockoala.com\/index.php\/wp-json\/wp\/v2\/comments?post=240"}],"version-history":[{"count":0,"href":"https:\/\/blockoala.com\/index.php\/wp-json\/wp\/v2\/posts\/240\/revisions"}],"wp:attachment":[{"href":"https:\/\/blockoala.com\/index.php\/wp-json\/wp\/v2\/media?parent=240"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blockoala.com\/index.php\/wp-json\/wp\/v2\/categories?post=240"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blockoala.com\/index.php\/wp-json\/wp\/v2\/tags?post=240"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}