Agentic SAP Business AI architectures implement intelligent enterprises with advanced generative AI capabilities on business technology platforms like SAP BTP, Azure or AWS.
SAP Business Agents offer new ways to implement AI-driven business processes and interoperability with other Cloud ERP systems or SaaS applications.
Embedded into the SAP Business Suite, AI agents enhance capabilities of SAP SaaS applications in financial, spend, supply chain, customer experience and human capital domains.
LLM-based AI agents demonstrate exceptional performance across a broad spectrum of business tasks with SAP business domain specific knowledge. Data-driven SAP AI agents create value with insights and help to perform core SAP business processes efficiently.
SAP Business AI Agents can be classified into different levels, with categories like performance, characteristics, domain knowledge or interoperability.
Basic responding SAP Business AI agents process inputs with reactive actions like multi-modal information retrieval or content generation.
Higher level SAP Business AI agents extend these basic capabilities with actions to call external tools, APIs or functions. These Business AI agents implement client side capabilities of the Model Context Protocol (MCP) protocol to call external data sources or tools.
Multi-agent SAP business processes are realized with different domain-specific business AI agents which have to be orchestrated by supervisor agrnts or orchestrators like SAP Joule to perform end-to-end workflow steps autonomously.
AI-powered assistants or agents empower SAP Business Suite with access to business context across domains, systems and organizations. To orchestrate the interaction in SAP multi-agent business AI scenarios, AI agents have to provide interoperability capabilities which can be implemented with standardized protocols like A2A (Agent2Agent) or MCP (Model Context Provider). A2A and MCP can be combined to extend the model context of SAP Business agents with external calls in multi-agent Business AI scenarios.
The Model Context Provider (MCP) protocol implements a client-server architecture with json-rpc, server-side events (SSE) and streaming http communication technologies.
MCP servers act as shared memory or context layer, offer tool catalogs and implement connectivity to MCP clients. MCP Client LLMs compare user queries against the tool descriptions to select the best fit tool for execution.
MCP can be combined with RAG to augment the LLM context with information retrieved from external systems or with GenAI components which generate new content.
MCP clients are powered by agentic LLMs and can be realized with SDK like Langchain (Agent Executor) or SAP Generative AI Hub. FastMCP is a popular Python framework to implement MCP clients and servers for agentic business applications.
Multi-agent SAP architectures integrate SAP and non-SAP business AI agents across different cloud platforms with Agent2Agent (A2A) interoperability.
Business AI agent registries and catalogs on multiple cloud platforms can advertise agent card information used by A2A clients to connect and collaborate with registered business agents. Empowered by orchestration, collaboration and interoperability, multi-agentic SAP business AI solutions work autonomously across business domains.
The amount of AI Agent Builder platforms is rapidly growing on all cloud environments. These platforms offer capabilities to create, manage and consume AI agents and examples are GCP Vertex AI, Joule on SAP BTP, AWS Bedrock or Azure AI Foundry.
Agent builder integrate frameworks like LangChain, LangGraph or Crew AI with multi-step reasoning capabilities of LLMs, RAG knowledge bases, MCP tool invocation and memory management for context or session management.