At SAP Sapphire in May 2026, SAP made its biggest AI announcement so far: the Autonomous Enterprise, a vision in which AI agents run large parts of business processes and people direct the outcome. Behind the headline there are concrete products, some available now and some rolling out through the end of 2026.
What SAP announced
- Joule Assistants and agents. More than 50 domain-specific Joule Assistants orchestrating over 200 specialised agents across finance, spend and procurement, supply chain, HR and customer experience, rolling out through the end of 2026.
- Joule Work. A new workspace where users describe what they want to achieve in natural language, and Joule coordinates insights, workflows and agents across SAP and non-SAP systems. The mobile app is available; the desktop app is planned for general availability in the second half of 2026.
- SAP Business AI Platform. One foundation that brings together SAP BTP, SAP Business Data Cloud and SAP Business AI, with a knowledge graph and Joule Studio for building agents with no-code and pro-code tools.
- Claude inside SAP. SAP and Anthropic expanded their partnership: Claude becomes a primary reasoning and agentic capability across SAP's AI-enabled portfolio, connecting to systems such as S/4HANA, SuccessFactors and Ariba.
- Open agent ecosystem. Bidirectional agent-to-agent (A2A) interoperability, with Google Cloud and Microsoft among the first partners, planned for general availability in Q4 2026, plus a EUR 100 million fund for partners building agents.
What changes in practice
Until now, AI in SAP mostly meant assistance: summarising, searching, drafting. Agents go one step further: they take action inside a process, such as adjusting an order, triggering a workflow or preparing a closing entry. SAP's own framing is important here: when AI acts inside a customer's environment, it should do so "within the same controls that govern human decisions".
Three things to prepare now
- A clean core. Agents work through released APIs and standard processes. Heavily modified systems are harder to automate safely, which makes clean core a prerequisite for AI, not just for upgrades.
- Trusted data. Agents are only as good as the data and business context they can reach. Data quality and a clear data strategy matter more than ever.
- Well-defined processes with clear exceptions. The best first candidates are high-volume processes with explicit rules and a clear point where a human decides, such as approvals, confirmations or service requests.
Where to start
You do not need to wait for every agent to be generally available. Pick one process with measurable manual effort, define what "good" looks like (automation rate, accuracy, time saved), and run a controlled pilot. That builds the governance, data and experience you will need when agents scale across the business.