At SAP Sapphire in May 2026, SAP announced more than 200 specialised AI agents and more than 50 Joule Assistants, alongside plans to embed Anthropic’s Claude as a primary reasoning and agentic capability across SAP’s AI-enabled portfolio. The announcement got plenty of coverage. Most of it focused on the wrong question. The right one is whether your business is in a position to use any of it.
What the autonomous enterprise really means
SAP has introduced the concept of the Autonomous Enterprise: a model where AI agents execute tasks directly rather than just assisting the user, inside the same systems that govern your operations.
SAP describes Joule as the orchestration layer for business outcomes, with Joule Assistants and agents coordinating workflows, data and actions across S/4HANA, SuccessFactors, Ariba and connected systems. Claude adds reasoning and agentic capability through the SAP Business AI Platform, with Model Context Protocol (MCP) supporting connectivity across SAP and non-SAP environments and reducing reliance on bespoke point-to-point integrations.
SAP reported that LC Waikiki saw a 70% increase in operational efficiency and a 50% reduction in manual errors after deploying early Joule capabilities. A purchase order query that previously took up to 10 minutes now completes in roughly 3 seconds.
But a result like that does not materialise because SAP released new software. It materialises because the business had the data quality, process standardisation, and governance in place for agents to act on reliably. That prerequisite does not appear in SAP’s keynote.
Why SAP chose Anthropic
SAP’s decision to bring Anthropic into the Business AI Platform is a deliberate signal about the kind of reasoning and governance it wants around agentic AI. Anthropic built Claude with safety, interpretability, and controllability as core design properties rather than retrofitted features. In environments where an AI agent is closing financial periods, approving purchase orders, or triggering procurement changes, those properties have direct commercial and regulatory weight.
Anthropic co-founder Daniela Amodei framed it at Sapphire:
That framing names the risk in agentic AI: an agent acting on a wrong conclusion inside a live business process, with no clear audit trail and no straightforward way to reverse what it did. That is a different category of problem from a chatbot giving a bad answer.
Anthropic’s focus on those properties is especially relevant because SAP and Anthropic are building industry-specific agents for public sector, healthcare, education, life sciences and utilities, where regulatory complexity makes governance non-negotiable.
The clean core nobody is talking about
The Autonomous Enterprise depends on what SAP calls a clean core: a cloud ERP environment with accurate data, standardised processes, and minimal legacy customisation. This is not a footnote. It is the hard requirement underneath everything else.
Agents built on SAP operate on assumptions about how business processes are structured. If your environment has accumulated years of custom modifications, inconsistent master data, or workarounds baked into the process layer, agents will fail in ways that are unpredictable and difficult to audit.
SAP’s announcement describes what is possible at the top of the maturity curve. Most mid-market SAP customers are not there yet. That does not make the announcement irrelevant, but it does make the sequence matter. A business on ByDesign, or running a heavily customised S/4HANA environment, still has the preparation work ahead of it before it gets there.
What your business should be doing now
If you are already on SAP Cloud ERP (S/4HANA) with a clean-core deployment, the questions are practical: which processes are candidates for autonomous execution, and where does the distinction between agent recommendation and agent action matter for compliance? If you are on ByDesign or a complex on-premise environment, SAP is explicit that agentic AI adoption requires process redesign and change management alongside technical configuration. Speed of adoption is not the differentiator here; what counts is arriving at the platform in good shape.
How Codestone approaches this
Codestone is recognised as a leading SAP Cloud ERP partner in EMEA North and the UK, with 27 years of SAP implementation experience and managed services that keep ERP environments clean long after go-live. We run structured discovery engagements that assess your data readiness, process standardisation, and migration options before you commit to an implementation.
Frequently asked questions
What is the SAP Autonomous Enterprise?
The Autonomous Enterprise is SAP’s strategic vision for AI-powered operations, announced at SAP Sapphire 2026. Rather than AI tools that assist users, it involves AI agents that execute business tasks directly inside SAP systems across finance, HR, procurement, and supply chain. The framework runs on the SAP Business AI Platform, with Anthropic’s Claude as a primary reasoning and agentic capability.
What is a clean core and why does it matter for SAP AI agents?
A clean core is a cloud ERP environment with standardised processes, accurate master data, and minimal custom modifications. SAP AI agents are built on assumptions about standard process structures. Without a clean core, agents operate on unreliable foundations and produce unpredictable outcomes that are difficult to audit or reverse.
What should we do if we’re on SAP Business ByDesign?
SAP removed Business ByDesign from its price list for net-new customers in April 2026. Existing customers continue to receive support, compliance and maintenance updates, but the strategic direction is now SAP Cloud ERP. The agentic capabilities announced at Sapphire are built for SAP Cloud ERP (S/4HANA), so accessing them means migrating to the cloud platform first, with the data quality and process standardisation to support agentic workflows. A structured discovery engagement is the right starting point.
