| AI-native | Architecture where AI is designed into the core experience, process, foundation, and platform layers, rather than added as isolated features. |
| AI-first | AI capabilities embedded into existing products or workflows, but still bounded by application, data, and governance silos. |
| Autonomous Enterprise | SAP's vision for enterprises that use AI assistants, agents, connected data, and governed automation to operate across business domains with less manual coordination. |
| System of record | The transactional foundation that stores business facts, enforces rules, and records what happened. |
| System of context | An intelligence layer that connects data, process knowledge, decision history, interactions, and semantics so agents can reason with business context. |
| Cognitive Core | The intelligence foundation formed by business data, semantic knowledge, reasoning models, agents, and platform services. |
| Context engineering | The practice of assembling the right authorized, current, and relevant context for an AI interaction. |
| Harness engineering | The practice of wrapping models and agents with enterprise controls such as identity, policy, sandboxing, memory, observability, evaluation, and governance. |
| Specification engineering | The practice of defining agent behavior, constraints, success criteria, and evaluation measures precisely enough to test and improve outcomes. |
| Agentic orchestration | Coordination of agents, tools, APIs, events, data products, and workflows to decompose goals, act, observe results, and adapt. |
| Managed agent runtime | A platform-managed environment for deploying and running agents with sandboxing, identity, memory, observability, evaluation, and tenant isolation. |
| Agent identity | A first-class digital identity that lets an agent authenticate, receive scoped authorizations, act under policy, and be audited. |
| Data flywheel | A self-reinforcing loop where richer data improves AI models and decisions, AI-driven decisions improve processes, and improved processes generate even richer data. |
| Federated Knowledge Graph | A model where domain-specific knowledge graphs retain autonomy while connecting through shared enterprise concepts. |
| Semantic grounding | Anchoring AI outputs in metadata, domain models, data definitions, authorization rules, and enterprise semantics to improve accuracy. |
| Sovereign AI | AI architecture and operations designed around data residency, regulatory, infrastructure, model, and operational sovereignty requirements. |
| Human-in-the-loop | A control pattern where humans review, approve, override, or guide AI actions for high-risk or high-impact decisions. |