AI Agents That Execute Work, Not Just Assist

The company also unveiled SAP Autonomous Suite, which deploys more than 50 domain-specific Joule Assistants and more than 200 specialized AI agents. Unlike traditional AI copilots that primarily surface recommendations, SAP's agents can execute operational workflows directly. The Autonomous Close Assistant can automate journal entries, reconciliation and error resolution during financial close cycles — compressing what SAP claims can be a weeks-long process into days.

SAP also launched Industry AI, introducing eight autonomous industry solutions embedding sector-specific logic, regulatory requirements and operational data models into AI workflows. At Sapphire, SAP highlighted work with European energy giant RWE, where AI agents analyze offshore wind turbine incidents, identify likely root causes and generate prefilled maintenance work orders using historical operational data. "What makes us different from others is depth," Klein says. "Both horizontal across finance, supply chain, procurement, HR and CX, and vertical into 26 industries with domain-specific logic and regulatory expertise."

However, a substantial update involved Joule Work, SAP's new conversational user experience layer. Instead of navigating separate ERP applications and dashboards, users interact directly with Joule by describing business outcomes they want completed. The system then orchestrates workflows, data and AI agents behind the scenes. Klein asserts that Joule will increasingly become the main interface for business users as "people will focus on outcomes, not screens."

Klein argues governance becomes even more important as autonomy expands. "Every action an agent takes in our Autonomous Suite is fully logged," he explains. "You always know what an agent did, why it did it and what data it used." He describes the approach as "traceability by design" — transparency built into the system rather than bolted on as a feature.

To back that claim at enterprise scale, the company unveiled a partnership arsenal spanning much of the modern AI infrastructure stack.

Anthropic’s Claude will power Joule agents across HR, procurement and supply chain, grounding frontier AI in trusted business data and process context. NVIDIA’s OpenShell runtime is being embedded directly into SAP’s Business AI Platform to govern how those agents execute securely. Likewise, Amazon Web Services is building zero-copy integration between Amazon Athena and SAP Business Data Cloud, eliminating the replication bottlenecks that have historically slowed enterprise analytics.

Moreover, Microsoft is enabling bidirectional agent-to-agent communication between Joule and its own agent frameworks while expanding sovereign cloud support on Azure for customers with strict data residency requirements.

Palantir Technologies, meanwhile, is tackling the hardest migration scenarios — the complex, data-heavy transformations that have historically stalled cloud ERP projects — with Accenture serving as a co-innovation partner. Likewise, Mistral AI and Cohere bring sovereign model options for enterprises unwilling or unable to route sensitive workloads through American hyperscalers.

The Enterprise AI Orchestration Wars Have Begun

Nearly every major enterprise software company now wants to become the orchestration system through which AI agents reason, act and automate work. But each vendor approaches the problem from a different starting point.

Salesforce represents SAP's most aggressive near-term challenger. Agentforce initially focused on customer-facing automation but has expanded into operational workflows traditionally dominated by ERP vendors, including back-office automation across onboarding, auditing and enterprise workflows. Oracle remains SAP's most dangerous direct ERP competitor, with its Fusion Agentic Apps strategy embedding autonomous agents into procurement, finance and supply chain. Oracle's vertical integration — spanning infrastructure, databases, cloud platforms and enterprise applications — lets it pitch CIOs on fewer integration points and single-vendor accountability. But that same strategy creates lock-in concerns for enterprises trying to maintain model flexibility.

SAP is positioning itself deliberately against that closed-stack approach. "We don't want to own the front door by locking people in," Klein tells me. "Rather, earn it by being the most valuable layer in the stack."

Microsoft's advantage comes from ubiquity — Copilot, Azure AI and Copilot Studio increasingly control the productivity layer where employees already spend most of their time. SAP's interoperability announcements suggest coexistence is more realistic than displacement — a battle not over replacement, but over which layer becomes the primary orchestration surface.

ServiceNow presents another important rival around workflow governance. Both companies argue enterprise AI succeeds only when grounded in governed workflows and trusted operational data. Klein claims SAP maintains an advantage in deeply transactional financial environments. "In areas like finance, procurement and HR, our agents are developed to be fully audit-ready. That's fundamentally different from deploying a general-purpose AI and hoping it gets compliance right."

Can SAP Reclaim The AI Narrative?

SAP's stock reached an all-time high of $306.60 in July 2025 before pulling back sharply. Following Q1 2026 earnings, shares dipped more than 6% despite cloud revenue growing 27% year over year. Current cloud backlog reached €21.9 billion, up 25% at constant currencies, while Cloud ERP Suite revenue grew 30% year over year. For full-year 2026, SAP projects €25.8 to €26.2 billion in cloud revenue alongside roughly €10 billion in free cash flow.

Five years from now, Klein believes SAP's moat will come from trusted operational data, embedded process logic and governance infrastructure — not AI models themselves. "The data will matter because it's semantically rich and trusted," he says. "The governance layer will matter because regulation is only increasing. The applications will matter because they encode decades of process logic that no foundation model can learn from public data alone."