Sapphire 2026: what SAP's autonomous enterprise means for your landscape

At Sapphire in May 2026, SAP stopped describing AI as a feature and started describing it as the way its software will be used. The phrase it chose was the autonomous enterprise: business processes carried out by AI agents, with people setting goals, approving exceptions and staying accountable. For SAP customers the announcements matter less as a vision than as a set of products with dates, licences and prerequisites. This article separates the two.

What SAP announced

The SAP Autonomous Suite. SAP says it now has more than 50 domain-specific Joule assistants across finance, supply chain, procurement, HR and customer engagement. Those assistants coordinate a larger set of specialised agents, which SAP put at more than 200. Some agents respond when a user asks; others run in the background on business events or schedules.

Joule as the front door. Joule, first launched in 2023 as a copilot, is being repositioned as the main engagement layer. Users ask for an outcome, Joule routes the request to the right assistant, and power users can call a specific agent directly. SAP also showed Joule Work, an interface where people describe what they want done and Joule orchestrates the workflows, data and agents involved, across SAP and non-SAP systems.

The SAP Business AI Platform. SAP Business Technology Platform, SAP Business Data Cloud and SAP's AI services are being brought together as one governed platform. The SAP Knowledge Graph gives agents a structured map of business entities and relationships, and a "company memory" layer is meant to capture policies, approvals and how exceptions were handled before.

Joule Studio 2.0. The tool for building your own agents now supports pro-code development in Python and TypeScript, including with coding assistants such as Claude Code and Cursor, with deployment to a managed runtime. Reports from the event said first customers would receive it from June 2026 and that it would be free through the end of the year. Agents built there are to support the open agent-to-agent (A2A) protocol.

Partners. Anthropic's Claude models were named as powering Joule agents in finance, procurement and supply chain. SAP also highlighted partnerships with AWS on data and with Palantir and Accenture on complex data migrations.

What it does not change

Agents act on the same data and authorisations as everything else in your system. An agent that proposes a dunning action is only as good as the customer master, the payment history and the dispute records behind it. An agent that books an accrual needs the same controls a person would. SAP says every agent action is logged and that it designed the framework with SOX auditors in mind, but the design of approvals, segregation of duties and evidence is still yours.

Analysts at the event were notably cautious. The common advice was to start with a small number of governed, use-case-specific agents, not hundreds, and to expect comfort with autonomy to vary by process. Finance leaders may accept an autonomous accrual long before they accept an autonomous close.

Where you are probably starting from

What you can switch on depends on what you run:

• SAP cloud applications (S/4HANA Cloud, SuccessFactors, Ariba, SAP Sales and Service Cloud) get the broadest coverage first.
• RISE with SAP customers were told they would get three Joule assistants activated in their first year; GROW with SAP customers get access to the agent portfolio on onboarding.
• On-premise and older ERP customers will see some assistants and agents that work across their systems, according to SAP, but coverage will be narrower and timelines less specific.

A practical first step

Most of our clients get more from a short, structured assessment than from waiting for the full catalogue:

1. List the processes with obvious exception work. Collections, invoice matching, supplier onboarding, maintenance planning and order changes are typical.
2. Check which assistants and agents exist for your products and releases, and what licences or AI units they consume.
3. Look at the data and roles behind each candidate. If the master data is poor or nobody can say who approves what, fix that first. The agent will not.
4. Pilot one or two scenarios with a baseline. Measure cycle time, touches per document or recovered cash before and after.
5. Decide what to scale based on the numbers.

The technology will keep moving monthly. The preparation work, meaning data quality, clean roles, documented processes and a clean core, pays off whichever agents you end up using.