Agentic AI in SAP

As of 2026-05-29

Agentic AI is the frontier: AI that takes action, not just answers. Where Joule (M046) answers "what is revenue?", an agent pursues a goal ("investigate the EMEA margin drop, find the driver, draft the fix") across data + processes. Four properties: goal-directed, multi-step, tool-using, collaborative. SAP's Joule agents / Agent Builder ground these on BDC data (M031/M033). Grounding + governance ARE the safety story — an agent acting on bad data or beyond its authority is a liability; the governance spine (data products M033, least-privilege M081, privacy M082, lineage M077, zero-trust M090) is what makes agentic AI safe. The model layer (incl. Anthropic's Claude via SAP's partnership, §0b) is swappable; the governed foundation + action-authorization design are the durable SAP-specific consultant value. Map use cases to maturity honestly (§0i); human-in-the-loop + auditability are requirements. Scarcest, highest-value emerging skill.