A finance leader looks at an overdue invoice. The ERP confirms the fact: Payment is late, the supplier is on file, the contract is active.
What it cannot say is why this supplier keeps slipping, what resolved a similar dispute last time, or that the same supplier has a delayed shipment in logistics and a renegotiated contract in procurement at the same moment.
The reasoning behind enterprise decisions has stayed locked in human judgment, scattered across systems.
For 50 years, enterprise software has been an excellent system of record. Closing the reasoning gap on top of it is what enterprise AI was always meant to do.
From AI-first to AI-native
The first wave, the AI-first approach, added intelligence inside existing applications. A feature can summarize an invoice or suggest a journal entry, but it lives within one application and cannot see across the landscape. Three barriers keep it confined: It lacks business and process context, it sits on disconnected systems without a shared data model, and it lacks the governance to be accountable at scale.
Meanwhile, the pace of change is unforgiving. Agentic systems, new interaction models, and new ways of grounding AI in business data are arriving faster than most architectures can absorb. As SAP CEO Christian Klein noted this year at SAP Sapphire, 80% accuracy may suffice for consumer AI; it is nowhere near enough for the world’s most business-critical processes. Bolting more intelligence onto isolated applications will not close that gap. It only multiplies the silos.
So what does it actually take to move beyond isolated AI features and build an enterprise that reasons, learns, and acts as one, without sacrificing the trust, governance, and reliability the business depends on? It is the question CIOs, CTOs, and enterprise architects are working through right now.
The foundation behind the Autonomous Enterprise
It takes a new foundation, and that is exactly what SAP’s AI-Native North Star Architecture provides.
This is not a white paper that sits on a shelf; it is the technology foundation SAP is actively building to bring the Autonomous Enterprise to life: a business where agents, orchestration, and data work in one continuous loop to turn intent into trusted outcomes.
The shift it enables is from AI-first to AI-native, where software operates across the landscape as a system of context: an intelligence layer connecting data, process knowledge, decision history, and semantics. Agents reason over the whole picture, not fragments. Every interaction feeds intelligence. Every correction becomes a learning signal. Value shifts from software as a service to outcome as a service.
AI-native paves the way for the Autonomous Enterprise: one system of context that understands disputes in service, delays in logistics, and contract changes in procurement all at once, and can act on them with full governance and accountability.
Philipp Herzig, CTO and Member of the Extended Board, SAP SE
Crucially, AI-native does not replace what already works. It pairs two complementary paths. The deterministic path keeps the predictable, rule-based execution that compliance depends on. The probabilistic, AI-native path adds reasoning that learns from data and experience. One is reliable but rigid. The other is powerful, but without context and control, often confidently wrong. Context engineering, guardrails, and observability bind the two, turning raw capability into reasoning the enterprise can trust.
The architecture delivers this through four reimagined layers that together form a cognitive core:
- The user experience layer shifts interaction from navigating apps to stating intent, with Joule as the central engagement point.
- The process layer turns applications into capability providers that expose stable APIs, events, and data for agents to orchestrate.
- The foundation layer is where data and AI come together as the intelligent core: orchestration, reasoning, and model services on one side; SAP Business Data Cloud and the SAP Knowledge Graph on the other, with SAP-trained models, including SAP-RPT-1 for structured business data, sitting alongside leading third-party models in one governed generative AI hub.
- The platform layer provides the runtime, governance, and harness that turn stateless models into reliable enterprise agents.
It defines the cornerstone architectural building blocks for agentic systems across experience, process, data, and platform, turning SAP’s unique business context into a living system of intelligence
What does this look like in practice? A finance analyst asks Joule to resolve high-value disputes likely to delay payment. Joule does not act alone. It coordinates AI assistants, which in turn direct specialist AI agents through agentic orchestration: the assistant decomposes the goal, delegates to a finance agent and a service agent, and reconciles their results. People set direction; assistants coordinate; agents execute. Those agents draw on the right information through context engineering, find the correct data through semantic grounding in SAP Knowledge Graph, and act within governed boundaries, routing only exceptions to a human. Each resolution becomes a decision trace that makes the next one smarter.
This is not theoretical. During the 2026 keynote at SAP Sapphire, SAP COO Sebastian Steinhaeuser pointed to life sciences customer Takeda, which is achieving up to 10% productivity gains, up to 25% reduction in revenue loss from stock-outs, and up to five percent reduction in safety stock through autonomous regulated manufacturing. That is what AI-native looks like at work.
Data was the moat of the last decade.
Context is the moat of the next.
Frontier models are available to everyone. Business context is not. Each resolved dispute, each corrected decision, each completed process adds to it, compounding with every interaction.
Trust is engineered in, not bolted on. A set of cross-cutting, SAP-managed qualities holds the layers together: integration, identity, security, observability, and extensibility, with resilience, compliance, and sustainability handled by the platform.
Autonomy only creates value when it is governed, so agents become first-class principals with their own agent identity, scoped to a bounded subset of permissions and audited like any enterprise actor. Harness engineering wraps each model with the sandboxing, memory, and guardrails that make it dependable.
As the paper puts it, the model reasons but the harness governs, and it is the harness, not the model, that determines the ceiling. Open standards such as the Model Context Protocol and Agent2Agent protocol let agents interoperate across the enterprise, while sovereign cloud options keep data residency and compliance built in.
This direction is being shaped with the customer community, not handed down to it: the architecture carries forewords from the leaders of the German-Speaking SAP User Group (DSAG) and Americas’ SAP Users’ Group (ASUG) alongside SAP’s own.
The North Star is a living document. Published openly on SAP Architecture Center, it will keep evolving as the technology and the agentic ecosystem advance, and as customer feedback shapes the design. If you build with SAP or build on SAP, this is your invitation: Read the architecture, push back where it should be sharper, and contribute. The same invitation extends to the wider SAP Architecture Center site, where SAP’s reference architectures are being built openly with the community.
Read the AI-Native North Star Architecture and open the full paper on SAP Architecture Center or download it as PDF.
Beyond the architecture itself is a single commitment: building systems that learn rather than dictate. For SAP customers, 50 years of process knowledge, governed data, and trusted decision frameworks compound into a new kind of enterprise intelligence that is reliable, transparent, and deeply human.
The Autonomous Enterprise will not arrive as a single product launch. It will be built layer by layer, decision by decision, on the foundation described here, one grounded interaction at a time.
Anirban Majumdar is head of the Office of the CTO at SAP.
PVN PavanKumar is vice president of the Office of the CTO at SAP.
