SAP: Enterprise AI Agents Require Knowledge Graphs and Governance

At VB Transform 2026, SAP outlined its vision for moving beyond basic chatbots to sophisticated, autonomous AI agents capable of executing business processes. The company emphasized that the key differentiator lies in grounding these agents in an enterprise's unique context rather than relying on general knowledge.

According to Max McPhee, senior solution advisor at SAP, AI agents begin to exhibit more advanced behavior, feeling like a "coworker" instead of an "assistant," when they are provided with context specific to the enterprise. This essential company context is effectively built using knowledge graphs and vector-embedded data, which facilitate easy information retrieval for agents and help them understand internal jargon and acronyms.

SAP draws parallels between onboarding AI agents and onboarding new employees. Beyond knowledge graphs, strong governance and process control — areas where SAP has extensive experience — are crucial. The company is modernizing its 50-year history in process management to accommodate the increased flexibility of AI agents.

Furthermore, SAP is leveraging machine learning to validate agent behavior, detect anomalies, and implement guardrails. Identity and permissions are critical security components. Both human users and SAP's AI assistant, Joule, must possess the necessary rights to access a system, preventing agents from bypassing access controls.

SAP is also focused on integrating its core knowledge with customer-specific landscapes. Acquisitions like LeanIX and Signavio, along with investments in automation company n8n, aim to map complex enterprise architectures. However, McPhee cautioned that modernization of legacy on-premises systems is necessary to avoid performance bottlenecks as autonomous agents are deployed more widely.