SAP launches its most ambitious AI platform yet in Madrid. With 200+ autonomous agents and a EUR 100 million ecosystem bet, and a clear answer on liability: not us.

More than 9,200 customers, partners, analysts and journalists descended on Madrid’s IFEMA congress centre this week for SAP Sapphire 2026, the European edition of SAP’s annual flagship event. The show was sold out, CEO Christian Klein noted during the opening keynote on Wednesday. The theme was unambiguous: SAP is no longer positioning itself as an enterprise software company. It is becoming, in Klein’s words, a business AI company.

The centrepiece announcement was the launch of the SAP Business AI Platform. A new integrated architecture that weaves together the company’s AI foundation, its Business Data Cloud and its Business Technology Platform into a single layer. It is designed to give AI agents direct access to ERP data, process logic and governance controls.

Klein framed the problem squarely at the start of his keynote: the large language models powering most enterprise AI tools today have never been trained on actual business data, do not understand company-specific governance requirements and cannot reliably navigate the kind of mission-critical, deterministic processes that ERP systems run.

“If agents run your payroll, your financial flows, your demand and supply chain planning, 80 percent accuracy is just not good enough,” he said.

Fifty years of ERP, now wired for AI

The solution SAP presented is built around what it calls a context layer. In which a knowledge graph maps AI queries to the right processes and data fields across an ERP landscape that spans more than seven million data fields.

Alongside that, SAP introduced new SAP Domain Models: foundation models trained on SAP code, data structures and business process documentation. Designed to ground AI responses in enterprise context rather than generic internet knowledge.

The Business AI Platform also incorporates the renamed Joule Studio 2.0. An open development environment for building AI agents that works across SAP and third-party LLMs. And a new AI Agent Hub for governance and transparency across the full agent landscape.

Philipp Herzig, SAP’s Chief Technology Officer, was characteristically direct about why this took as long as it did. “The most difficult part was that we had all these elements, but they were not connected,” Herzig said.

The breakthrough, Klein added, came roughly eight to nine months ago, when the company had what he described as a “co-web moment”: the data layer, the development environment and the governance layer had to be engineered as a single system, not assembled from separate products.

The autonomous suite: 200+ agents, one interface

On top of the SAP Business AI Platform sits what SAP is calling the Autonomous Suite: more than 50 domain-specific Joule Assistants spanning finance, supply chain, procurement, HR and customer experience. With each orchestrating a subset of over 200 specialised agents to execute precise tasks.

The most concrete example on stage was the Autonomous Close Assistant. SAP says it can compress the financial close process from weeks to days. By automating journal entries, reconciliation and error resolution end to end.

SAP also announced Joule Work, a new user interface layer that replaces navigation between applications with a conversational interaction model: a user describes a desired business outcome and Joule orchestrates the workflows, agents and data to deliver it, across both desktop, mobile and voice.

The Autonomous Suite extends into industry-specific territory through what SAP is calling Industry AI: seven autonomous solutions embedding sector-specific process logic and regulatory requirements.

At Sapphire, SAP highlighted its work with German energy company RWE. Where AI agents analyse historical incident data from offshore wind turbines to identify root causes and generate pre-filled work orders. Thus, saving 90 minutes per turbine repair.

Promising, but the maturity isn’t there yet

Customer reaction on the show floor ranged from enthusiasm to measured realism. Kenny van Sleuwen, System Architect at Dutch shipbuilder Damen Shipyards Group, called the knowledge graph announcement the missing piece of the puzzle: what would finally make it possible to navigate from a specific prompt to the right reasoning and context-aware response.

His caveat was equally specific. Damen has been using Joule on S/4HANA but found the current version required highly precise prompting to deliver useful output. “The maturity is not yet there for it to reason across process and company context the way a real co-pilot would,” he said. With the knowledge graph, that confidence has arrived.

Erik Grave, Process Manager for Requisition-to-Pay at Dutch Energy Network Operator Gasunie, was more cautious. Gasunie went live on S/4HANA in October 2024 via a full greenfield migration and has spent the past year stabilising its procurement processes.

Grave said the concept of AI agents taking over procurement steps was interesting but that the human factor had to remain central. “The complexity of what is being ordered matters,” he said: “for straightforward purchases an agent can handle the checks, but human judgment cannot be removed from every step.”

He confirmed that Gasunie is actively exploring the use of agents for specific automation tasks in its procurement desk operations.

Sovereign by design, liable by choice

On sovereignty, SAP used Sapphire to announce that Mistral Plus is now generally available on the SAP Business AI Platform within its sovereign cloud infrastructure. And that Cohere North will follow in June. Martin Merz, SAP’s head of Sovereign Services and Delivery, explained the company’s approach as a spectrum of four dimensions: data sovereignty, operational sovereignty, technical sovereignty and legal sovereignty.

“It is not one-size-fits-all,” Merz said. “Military customers may be perfectly comfortable running on AWS, as long as it is approved by their cybersecurity agency. Others will only accept deployment in their own data centre. Sovereignty is personal.”

The question of who bears liability when an autonomous agent makes a wrong call surfaced repeatedly across sessions. Asked directly who is responsible when one of the 224 announced agents makes an erroneous decision in payroll or financial close, Sebastian Steinhaeuser, SAP’s Chief Operating Officer, did not answer the liability question directly. Instead, he pointed to the configurable human-in-the-loop controls built into the platform.

“Think of it like a dog on a leash,” he said. “You start with a short leash, with a human reviewing every decision. As trust builds and the agent proves it delivers accurate results, you let the leash out further. All the way to full autonomy if you choose.” Martin Merz, who was asked the same question in a separate conversation, was equally clear that SAP does not see itself as the liable party.

“It is always the customer who has the call,” he said: who decides what to do with the model, how much freedom to give it, what data it can access. “It is completely their decision,” he said.

€100 million and an open platform bet

On the ecosystem and partnership side, SAP announced a 100 million euro fund for partners to develop AI assistants and agents on the new platform. Joule Studio 2.0 will be available free of charge through the end of the year.

Confirmed platform partnerships include Anthropic, whose Claude models will power Joule agents across HR, procurement and supply chain; NVIDIA, whose OpenShell provides the secure runtime environment for Joule Studio; n8n, the German workflow automation unicorn, which will bring visual AI workflow orchestration inside Joule Studio; Mistral AI and Cohere for sovereign model options; and Google Cloud and Microsoft for bidirectional agent-to-agent interoperability with external frameworks.

SAP also used the event to announce acquisitions completed since its previous Sapphire: Dremio, for high-performance data lakehouse capabilities, and Reltio, for master data management, both now integrated into the SAP Business Data Cloud.

A partnership with Palantir and Accenture for complex data migration scenarios and a new agent-led ERP migration toolchain that SAP claims can reduce migration effort by more than 35 percent rounded out the transformation services announcements.

Cloud or legacy: SAP closes the gap

The customer journey program also receives an update. RISE with SAP will now activate three Joule Assistants within a customer’s first year,. With contractual commitment from SAP to deliver them live. SAP Grow customers get full portfolio access at onboarding.

Crucially, for the large installed base still running on-premise S/4HANA or SAP ECC, SAP announced that customers committing to move the majority of their landscape to SAP Cloud ERP will gain access to select AI scenarios. An explicit attempt to prevent a growing capability gap between cloud and legacy customers.

Klein opened his keynote with a provocative question: will SAP still be a software company in the future? He left the answer to Joule, who delivered it without hesitation at the close of the session. SAP is becoming a business AI company, the AI assistant declared. Whether customers are ready to become autonomous enterprises at the same speed is a question no AI assistant answered in Madrid.

The post SAP Sapphire 2026: The autonomous enterprise takes shape in Madrid appeared first on Enterprise Times.