AI agents are redefining enterprise software architecture, unlocking new levels of adaptability, intelligence, and human-machine collaboration. Mastering agentic AI systems will be critical for organisations seeking to drive growth, resilience, and innovation in an increasingly complex business landscape.
For decades, SAP has been the backbone of global business operations. That foundation is now evolving rapidly with an AI-first, suite-first strategy designed to embed intelligence at every layer of enterprise software. Rather than treating AI as a bolt-on capability, SAP is reshaping its entire application suite around embedded, contextual, and agent-driven intelligence.
Joule, SAP’s new orchestration layer for AI agents, is at the heart of this transition. Joule integrates applications, structured and unstructured data, and multi-modal AI capabilities into a dynamic platform. It acts as an AI-native user interface, a process conductor, and a planning engine capable of navigating uncertainty and complexity.
“The first misconception we address is that AI is a monolith,” Jared Coyle, Chief AI Officer North America, SAP America, says. “Organisations will consume AI in a few different ways: embedded within applications, through custom capabilities, and increasingly through multi-agent systems that plan, reason, and collaborate with humans and other agents.”
SAP’s strategy uses its unparalleled access to business data, accumulated across five decades of ERP leadership, as the raw material for this agentic evolution. With structured knowledge graph layers, retrieval-augmented generation (RAG) frameworks, and agentic runtimes, SAP offers a coherent environment where AI can reason, act, and adapt within the real constraints of enterprise systems.
From automation to orchestration
Historically, enterprise automation has focused on deterministic workflows and transparent, repeatable processes optimised through rules and conditions. The analogy is a train running on fixed tracks. AI agents introduce an entirely different paradigm: dynamic, self-planning systems more akin to self-driving cars, capable of navigating unpredictable terrain.
“Before, one business problem had one exact solution,” Jochen Schneider, Head of BTP AI at SAP, explains “Now, there can be multiple right solutions. Agents take a mission prompt, use tools, apply reasoning, and plan dynamically to reach outcomes. It is a fundamentally more flexible, resilient model of automation.”
In SAP’s architecture, an AI agent consists of four core building blocks: instructions (prompts), reasoning and planning capability via foundation models, access to enterprise tools, and memory to store short- and long-term learnings. This allows agents to shift from executing single tasks to orchestrating entire processes, adjusting plans in real time as data and conditions change.
The impact is not incremental. SAP projects efficiency increases of up to 30 per cent across domains such as financial management, supply chain, spend management, human capital, and customer experience. Combining semantically aligned SAP and non-SAP data products, Joule’s orchestration capabilities and embedded generative AI positions enterprises to move beyond task automation towards true business autonomy.
Real-world examples of agentic transformation
Theory only becomes valuable when it translates into practice. SAP provides concrete examples of how agentic AI transforms daily operations. In one scenario, a procurement professional tasked with sourcing consultancy services from framework contracts was used to manually extract rates from PDF documents, calculate costs, and match them with ERP records. Through SAP Build and the Project Agent Builder, an AI agent reads PDF contracts, calculates consultancy costs, searches for supplier information, and integrates directly with SAP S/4HANA.
The agent operates across multiple tools, document extraction, web search, coding, and automation workflows, dynamically reasoning about the best partner selection based on price, location, and contract terms. Human oversight remains critical, but the tedious manual processes have been delegated to the agent, allowing procurement professionals to focus on strategy rather than administration. This is not a hypothetical case. It is a live deployment with measurable impacts on time savings, error reduction, and employee satisfaction.
The architecture of enterprise AI agents
SAP categorises its agent offerings across three levels of complexity, all anchored to Joule’s orchestration layer.
At the simplest level are single-shot skills and Joule scenarios, specific tasks triggered conversationally by the user. As organisations mature, Joule orchestration allows chaining skills into more complex, adaptive workflows through dynamic planning and scenario dependency mapping.
Content-based agents allow user-defined scenarios, declarative tool use, and managed runtime environments. These agents are ideal for moderately complex, configurable business processes requiring flexible, structured execution.
Code-based agents are available for the most complex and mission-specific requirements. These offer developers complete control to script multi-LLM access, bespoke tool usage, and custom orchestration logic, all within the enterprise security and governance frameworks expected from SAP.
The underlying architecture ensures that whether enterprises build light-touch agents or complex autonomous workflows, they operate cohesively, leveraging the same semantic data models, security protocols, and orchestration runtimes.
Joule native agenticness and the rise of dynamic planning
One of the defining innovations in SAP’s approach is ‘native agenticness’ within Joule. Agents operating in Joule can collect relevant grounding knowledge dynamically, plan based on LLM-intrinsic and external data, execute those plans and loop back to re-plan if real-world execution diverges from expectations.
This built-in capability to reason and re-plan on the fly is vital. In real enterprise environments, unexpected changes are the norm rather than the exception, and static workflows break under this volatility. With dynamic grounding and planning loops, Joule’s agentic runtime brings the resilience needed to support mission-critical processes in finance, supply chain, and customer operations.
As Coyle describes, “We move from thinking about regulation and compliance as limiting factors to using AI to innovate and adapt within compliant, auditable frameworks,” he adds. “The ability to dynamically ground, plan, and execute changes everything about how enterprises can build resilience and agility.”
The human-machine partnership
Despite the sophistication of agentic systems, SAP maintains a clear philosophical stance: agents are designed to augment human capabilities, not replace them.
The concept of human-in-the-loop remains fundamental, particularly in high-stakes decision-making contexts such as financial planning, procurement negotiations, and workforce management. Human oversight ensures that ethical considerations, corporate responsibility, and strategic judgment remain embedded in decision processes.
“We do not build agents to replace professionals,” Schneider explains. “We build them to amplify expertise, automate the mundane, and free up human intelligence for higher-order problem solving. This is how we unlock new levels of value and innovation.”
Governance, security, and traceability are embedded in every aspect of the agentic platform. Each action, decision point, and outcome can be audited, ensuring enterprises can scale agentic operations without sacrificing control or accountability.
Unlocking a new era of enterprise innovation
The real promise of SAP Business AI and agentic architectures is not simply cost reduction or operational efficiency. It is about unlocking entirely new ways of running businesses, which will be more adaptive, intelligent, and human-centric.
Integrating AI agents into the core of enterprise systems represents the next evolutionary leap in automation. It moves organisations from linear, pre-programmed workflows to dynamic, goal-oriented orchestration across complex, changing environments.
As Fei-Fei Li eloquently captures, “The future of artificial intelligence is not about man versus machine, but rather man with machine. Together, we can achieve unimaginable heights of innovation and progress.”
Those enterprises that embrace agentic thinking today will not merely automate their past processes. They will reinvent what is possible, architecting self-driving organisations capable of thriving in an era of perpetual disruption.
