1) Specialized foundation models will outperform general-purpose LLMs in enterprise tasks

Large language models opened the door to many enterprise use cases, but they are not equally strong across all business needs, especially where accuracy, structured reasoning, and prediction matter. SAP highlights a shift toward specialized foundation models, including “relational” models trained on structured enterprise datasets. The point is straightforward: when the goal is forecasting delivery dates, spotting anomalies, or estimating risk with repeatable accuracy, purpose-built models should deliver better outcomes than generic chat-first systems.

2) AI-native architecture becomes the new baseline

In 2026, the focus shifts from “adding AI features” to building AI-native applications—systems designed from the start to combine deterministic systems of record with a continuously learning intelligence layer. SAP emphasizes the role of enterprise context: semantically rich data foundations (often described as knowledge graphs) that make AI grounded, reliable, and improvable over time. The practical implication is that AI performance will be increasingly limited (or unlocked) by data connectivity, semantics, and architecture—not just by the model itself.

3) Agent governance becomes mission-critical

As AI agents evolve from answering questions to executing multi-step work (planning, tool use, collaboration, self-checking), enterprises face a familiar risk: uncontrolled growth—this time with autonomy, sensitive data, and real operational impact. SAP argues that governance must cover the full lifecycle: defining what agents can do, enforcing policies, ensuring auditability, monitoring performance, and setting clear human-in-the-loop boundaries. In other words, organizations will need to manage agents like a digital workforce—with rules, oversight, and accountability.

4) Intent-driven ERP and “generative UI” redefine the user experience

SAP expects user experience to shift from navigating applications to expressing intent. Instead of clicking through screens, users state outcomes—such as preparing a customer visit or summarizing purchasing performance—and an agent orchestrates the steps across systems, generating the necessary views, actions, and workflows. This is sometimes described as “generative UI” or even “no-app ERP.” The traditional backbone still matters—controls, security, master data, and business logic—but the front-end interaction becomes outcome-driven, faster, and more contextual.

5) Deglobalization accelerates sovereign AI requirements

Geopolitics and regulation are pushing sovereign AI from a niche concern into a mainstream requirement. This goes beyond data residency: enterprises increasingly care about where models run, how they are operated, who controls infrastructure, and what jurisdictions apply across the stack. SAP’s view is that customers will demand AI that remains state-of-the-art while meeting locality and compliance requirements—driving more region-aware architectures and deployment options.

6) What leaders should do now

SAP’s conclusion is clear: success in 2026 will go to organizations that build AI on solid foundations—connected and trustworthy data, AI-native architecture, governed agent adoption, and user experiences designed around intent and outcomes. The recurring constraint remains the same as it has always been in enterprise IT: poor data quality and fragmented processes will cap AI impact no matter how advanced the models become.

7) How con4PAS can help

For many organizations, the gap is not access to AI—it is the ability to scale it safely and profitably. con4PAS supports customers in translating these AI themes into executable roadmaps across SAP landscapes, with a focus on practical outcomes: automation that holds up in real operations, compliance-by-design, measurable productivity gains, and adoption that sticks. This is especially relevant where procurement, finance, and shared services need reliable orchestration across processes, systems, and stakeholders.

Source: SAP News Center, “AI in 2026: Five Defining Themes” (January 2026).