Explaining the Foundation Layer as Intelligent Core

Imagine a business scenario where an intelligent agent needs to analyze vast amounts of financial data, predict market trends, and recommend optimal investment strategies, all while adhering to strict compliance rules. This agent can't operate effectively with isolated data or generic AI models. It needs a robust core that provides deep business context, specialized reasoning, and continuous learning. This is precisely the role of the Foundation Layer in an AI-native architecture. It's the intelligent core that integrates orchestration, reasoning, model services, and data grounding, enabling enterprise AI to reason over business data with unparalleled accuracy and relevance.

The Foundation Layer serves as the intelligent core of enterprise processes, where data and artificial intelligence converge to provide a compounding advantage for SAP customers. Traditionally, data and intelligence often remained separate, with models running without business context and data residing in silos, limiting their reasoning capabilities. The shift to an AI-native model fundamentally redefines this, creating a layer where enterprise intelligence is built, grounded, and continuously improved.

In this AI-native paradigm, orchestration, reasoning, and model services power the AI side, while governed data and semantic grounding, supported by SAP Business Data Cloud and SAP Knowledge Graph, provide the essential context. Together, these components store and compound enterprise knowledge, which grows and refines with each interaction. This integration allows intelligent systems to operate natively within enterprise processes, complete with built-in guardrails for ethics, security, compliance, governance, and efficiency.

At the heart of the Foundation Layer's intelligence is orchestration, primarily driven by Joule. Joule acts as the central orchestrator for agents and tools, managing their lifecycle and aligning their actions with overarching workflow goals. Its core function involves planning, routing user requests to the most appropriate agent, and continuously applying reflection to evaluate progress, resolve conflicts, and re-plan as needed. This ensures that users receive not just fast answers, but trusted and accurate ones.

For complex tasks, Joule can decompose them into smaller sub-tasks, delegating these to domain-specific agents that can execute in parallel and report back for coordination. Each individual agent operates within its own agentic loop, invoking tools through the Agent Gateway for specific operations. A key innovation here is the increasing use of self-verification mechanisms within agents, employing internal feedback loops to check accuracy during execution. This significantly reduces the error buildup that has historically hindered the scaling of multi-step workflows.

SAP's five decades of specialized codebase and domain knowledge represent a significant strategic advantage that general-purpose models cannot replicate. Frontier models typically lack exposure to SAP's internal code, architectural patterns, or intricate business logic. To address this, SAP-trained models combine continuous pretraining on internal codebases with fine-tuning and in-context learning. This approach provides a deep understanding of how various business processes, such as procurement workflows, connect to finance, how supply chain exceptions trigger approvals, and how specific business rules operate in production environments.

These models are hosted and consumed through the Generative AI Hub in SAP AI Core, which provides a harmonized interface to multiple frontier models, including third-party and SAP-hosted options. This hub ensures flexibility by allowing customers to choose the right model for each task without code changes and offers fallback orchestration if a preferred model is unavailable. A notable example is SAP RPT-1, a relational pretrained transformer specifically designed for structured business data like financial ledgers and transaction records, delivering training-free predictions within agentic workflows.

To ensure consistency and reduce duplication across applications and agents, the Foundation Layer promotes reusable intelligence. Core cognitive capabilities, such as document understanding, tabular analysis, and visual inspection, are implemented once and then exposed as composable skills. This approach minimizes redundant compute and infrastructure usage, making these intelligent capabilities readily available as building blocks that agents can invoke for specific tasks. This standardization streamlines development and deployment, ensuring that common AI functions are consistently applied throughout the enterprise.

Reasoning without robust data is mere guesswork. The Foundation Layer addresses this by integrating SAP Business Data Cloud (BDC) and SAP Knowledge Graph to provide trusted, context-rich, and efficient data and business context for agents to reason, act, and learn. This creates a self-reinforcing flywheel: applications generate data, BDC curates it, the Knowledge Graph adds semantic meaning, AI reasons over it, and the results feed back into smarter applications.

SAP Business Data Cloud connects data from across the enterprise—including SAP cloud and on-premise systems, legacy systems, and third-party sources—into a single, governed landscape. At its core are Data Products:

Data Products

Curated, governed datasets with clear ownership, schema, authorization rules, and lifecycle management. These serve as the primary governed interface for all data consumers, from dashboards to AI agents.

Data products follow an authorization model and are accessible through multiple channels: SAP Analytics Cloud to create business dashboards and enhance planning, the SAP Datasphere solution for use in semantic data modeling, partner platforms such as those from Databricks Inc. and Snowflake Inc. for data science, and AI agents supported by Joule that access data through natural-language-to-SQL generation grounded in SAP Knowledge Graph. SAP Business Data Cloud will integrate tightly with the SAP AI Core foundation to train machine learning models and ground large language models, enabling the creation of new data products from the resulting insights.

To help ensure agents have data access immediately, major SAP application tenants are provisioned with an embedded data foundation that provides auto-generated data products. These data products are based on application metadata, removing the barrier of limited data product availability, so Joule and agents can query across application boundaries from day one.

The SAP Knowledge Graph is the semantic backbone of the AI-native architecture. It links natural language inputs to SAP’s structured metadata, significantly reducing hallucinations and unlocking access to thousands of APIs, data models, and business entities across SAP. For instance, when a user asks to "show me overdue orders," the Knowledge Graph automatically discovers the correct API, identifies filter parameters, and constructs an accurate query. It connects various layers of enterprise knowledge, including API metadata, business semantics, data product metadata, and customer-specific extensions, creating a unified, logical view across the SAP solution landscape.

The long-term vision is a customer-specific knowledge graph with a shared enterprise ontology, where domain-specific knowledge graphs maintain autonomy while connecting via shared concepts, presenting a single logical view across the SAP solution landscape.

Every time an agent is initialized, it possesses immense intelligence but lacks specific context. Context engineering is the discipline within the Foundation Layer that addresses this by assembling the precise slice of enterprise information relevant to each interaction. It filters out redundant, outdated, or unauthorized data, ensuring agents operate on context that is both relevant and proportionate to the task, thereby optimizing retrieval and inference calls.

The Foundation Layer also incorporates a robust memory persistence mechanism. Agents capture decision lineage—what was known, considered, chosen, and resulted—across interactions. Every human correction becomes a structured decision trace. These traces, combined with the SAP Knowledge Graph, form a context graph: a living representation of enterprise knowledge that connects semantics, decision history, and operational context. This continuous learning from experience allows the system to move beyond simply responding to requests and begin anticipating business actions, ultimately leading to a compounding competitive advantage where context becomes the decisive moat.

To maintain and improve agent quality, the Foundation Layer implements structured evaluation cycles. These cycles go beyond technical assessments of individual agents (e.g., goal completion, tool call efficiency) to include process KPIs and financial/operational outcomes. This ensures agents are measured not just on technical performance but on their actual business impact. Evaluation is treated as a multi-objective optimization problem, where agents refine their strategies across multiple constraints—such as resolution time, customer retention, and quality—rather than optimizing a single variable. The business KPIs serve as feedback signals, enabling agents to iteratively improve towards a stable optimum.

This continuous improvement, coupled with meta-learning capabilities, allows agents to refine their learning methods, transforming workflows into cumulative assets that grow stronger and smarter with each cycle.

• Intelligent Core: The Foundation Layer integrates orchestration, reasoning, model services, and data grounding to create enterprise intelligence, moving from isolated data to context-rich reasoning.
• Joule Orchestration: SAP Joule acts as the top-level orchestrator, managing agent lifecycles, decomposing tasks, and applying reflection for trusted, accurate outcomes, with agents executing on BTP Fabric.
• Domain-Specific Models: SAP leverages its deep business context to train specialized models hosted in the Generative AI Hub, providing domain-specific knowledge that generic models lack.
• Data and Semantic Grounding: SAP Business Data Cloud provides governed Data Products, while SAP Knowledge Graph offers semantic grounding, linking natural language to structured metadata and enabling agents to reason over comprehensive business context.
• Context Engineering and Memory: The layer assembles relevant context for each interaction and persists decision lineage through a "context graph," allowing agents to learn from experience and continuously improve their reasoning.