Joule, Skills, Agents, and Foundation Models: What SAP's AI Family Means Commercially
Joule is not a single product but a family of frontend, Skills, Agents, and Foundation Models. Each component carries its own licensing logic, consumption patterns, and governance moments. What separates them commercially, how to distinguish Skills from Agents, and which Foundation Models are accessible in the GenAI Hub.
Joule for End Users
When SAP projects refer to "Joule," they typically mean Joule for Users: the conversational assistant embedded directly in SAP solutions. As of Sapphire 2026, Joule for Users is active in 35 SAP solutions, including S/4HANA Cloud, SuccessFactors, Ariba, and Fieldglass. End users can initiate tasks, trigger workflows, and access embedded AI features through a unified interface, without switching between applications.
For contract governance, this component is frequently the first governance moment: it is included in many RISE contracts and Cloud ERP bundles, but the precise scope is not always clear. Access to Joule for Users is not a binary on/off. It depends on the contracted tier, the activated solution, and whether AI Unit allowances are available to cover actual usage.
Joule Work, the UX vision introduced at Sapphire 2026, takes this further. SAP positions Joule as the primary interaction layer long term, gradually replacing traditional application navigation. What functions today as a copilot feature is intended to become the standard interface for all SAP solutions over time. This directional shift is commercially relevant because it structurally changes usage intensity and therefore consumption volumes.
Joule for Developers and Consultants
The second variant targets not end users but developers and implementation partners.
Joule for Developers is the ABAP assistant in the ABAP Development Tools environment (ADT) and in Business Application Studio (BAS). It is built on SAP-ABAP-1, SAP's proprietary model trained on 250 million lines of ABAP code, and supports code generation, refactoring, and test generation. SAP cites a productivity improvement of 20 to 25 percent for developers (source: SAP Community, Our 2026 Roadmap for Joule for Developers). Licensing is via a separate Joule for Developers SKU or as part of BTP developer subscriptions.
A distinct governance moment lies here: many organizations have Joule for Developers as an embedded feature within BTP subscriptions, without having actively managed that access. As usage grows, so does the AI Unit consumption attributable to this component.
Joule for Consultants remains in beta status and is designed as a configuration assistant for implementation partners. The generally available version is planned during the course of 2026. For active projects, the timing of that status change is relevant: beta capabilities can be modified or withdrawn without prior notice.
Joule Studio 2.0 is the development environment for customer-built agents. It has been generally available since June 2026 and includes Agent Builder, Skill Library, Joule Studio CLI, VS Code Extension, and an MCP Server integration for external AI toolchains. Through end of 2026, SAP provides free design-time access under fair-use conditions. This covers the development environment, not production operations. Runtime licenses for customer-built agents running in production require separate planning and negotiation.
Skills Versus Agents: the Decisive Distinction
No distinction in SAP's AI family matters more commercially than the one between Skills and Agents. It determines consumption volumes, risk profile, and which governance moments apply.
A Joule Skill is a deterministic, predefined operation. It executes a single, clearly defined action: create a purchase order, verify a record, forward a notification. Skills are predictable in consumption, manageable in risk, and are typically already factored into sizing assumptions at contract signing. As of Q2 2026, SAP provides more than 2,500 pre-built Skills in the Skill Library.
A Joule Agent is an autonomous system that orchestrates multiple Skills, plans, and acts independently. It pursues a higher-level objective and selects its own path toward that objective. A typical agent task: "Optimize all open purchase orders by delivery time and cost" or "Onboard the new employee across all systems." The Agent decides which Skills to activate and in what sequence, and can take paths that were not fully anticipated in advance.
The commercial difference is substantial. Agentic deployments consume three to twenty times the AI Units compared to pure copilot projections, based on observations from practice (source: SAP Licensing Experts, Negotiating SAP AI Contracts). Organizations making the shift from copilot deployments to productive agents without adjusting the sizing factor expose themselves to unexpected overage.
For the governance moment at contract negotiation, this means: the question "Are we deploying Skills or Agents?" should be answered before sizing is fixed. SAP standard sizing assumptions are typically based on skill-level usage. Organizations planning agents should bring this factor explicitly into the contract baseline.
As of Sapphire 2026, the SAP Autonomous Suite offers 224 agents and 51 role-specific assistants across Finance, Spend Management, Supply Chain, HR, and Customer Experience. The number grows each quarter. Each new capability is a potential new consumption point that warrants continuous monitoring.
Foundation Models in the GenAI Hub
Joule has no "Joule model" of its own. It is a frontend layer that consumes Foundation Models from the Generative AI Hub. The model selected for a given use case affects both the outcome and the consumption, because token rates vary substantially across models.
The Generative AI Hub provides access to Foundation Models from major providers. SAP abstracts the respective API so that customers can select a model per use case without establishing direct contractual relationships with the model vendors. Available models vary and are updated by SAP on a rolling basis. Which models are concretely available in a customer tenant depends on region, contract version, and SAP release cycle (source: SAP Help Portal, Generative AI Hub).
Two aspects are relevant governance moments for ongoing operations.
First, model selection is a cost parameter. Tasks that can be handled by a lighter model generate less token consumption than tasks that require a complex model with a large context window. An undifferentiated configuration, where the same model is applied across all use cases, is frequently not economically optimal.
Second, model substitutability is a governance question. SAP can adjust the model portfolio in the Generative AI Hub on a rolling basis. Whether and how this affects the customer tenant depends on whether the Order Form contains explicit provisions on model continuity. The absence of such a provision typically means SAP can adjust the model portfolio unilaterally. Each model substitution cycle is a governance moment: use cases that depend on a specific model's behavior require re-validation when the model changes.
SAP-Proprietary Models and Vendor Lock-in
Beyond external Foundation Models, SAP develops two proprietary models optimized for specific SAP contexts.
SAP-ABAP-1 was trained on 250 million lines of ABAP code. It is the foundation for Joule for Developers and offers a measurable advantage over generic language models on ABAP tasks: higher code generation quality, more precise test suggestions, and deeper contextual understanding of SAP-specific constructs (source: SAP News, Joule Studio Enterprise Scale Agentic Development).
SAP-RPT-1 is trained on SAP business process data and enables process inference for embedded AI features. It supports predictions and recommendations along SAP standard processes.
The strategic question with SAP-proprietary models runs in two directions. They offer genuine value because they carry embedded SAP domain knowledge. At the same time, they create a dependency: organizations that align their ABAP development practices around SAP-ABAP-1 are building a practice that does not transfer easily to a different model (source: SAVIC Technologies, AI Foundation Architecture 2026).
This governance moment is strategic in nature: how much contract infrastructure and development practice should be explicitly built on SAP-proprietary models? And which clauses in the Order Form describe the conditions under which SAP may change or withdraw these models? These questions should be on the agenda in early conversations on AI governance.
SAP-proprietary models typically sit at the lower end of the token price scale relative to externally sourced Foundation Models. This makes them commercially attractive for frequently called, standardized use cases.
Multi-Agent Protocol and MCP
The sixth dimension of the Joule family concerns cross-system coordination. SAP introduced a multi-agent protocol in Q1 2026 that enables agents to coordinate across SAP and non-SAP systems.
In concrete terms: an SAP agent can communicate with another SAP agent, with a customer-built agent from Joule Studio, and with a third-party agent via the open Model Context Protocol (MCP). A hire-to-onboard workflow can coordinate a SuccessFactors agent for candidate selection, an S/4HANA agent for master data creation, and a further agent for system access provisioning, without manual handoffs between systems.
MCP is an open standard that SAP has supported since Q1 2026. It allows external AI toolchains to access SAP Joule Skills, and SAP agents to consume external MCP servers. For organizations operating their own AI architectures, this is a bridge between SAP agents and external tooling.
From a contract governance perspective, multi-agent scenarios open several governance moments simultaneously. Every agent call within a multi-agent workflow generates its own consumption. Visibility into these workflows is limited with standard SAP reporting tools. The SAP AI Agent Hub, planned for general availability in Q3 2026, is designed as the governance instrument for this complexity.
Also relevant is the tension between MCP openness and the SAP API Policy v4/2026: SAP supports MCP but simultaneously prohibits autonomous third-party AI agents at SAP APIs. Third-party components may provide Skills but may not operate independently acting agents against SAP data (source: SAP API Policy, cross-referenced in SAP Help Portal). Organizations planning hybrid architectures with SAP and non-SAP agents should clarify this boundary before the architecture decision is made.
FAQ
What is the difference between a Joule Skill and a Joule Agent?
A Joule Skill is a deterministic, predefined single action: predictable consumption, limited risk. A Joule Agent is an autonomous system that orchestrates multiple Skills and plans independently. The consumption difference is substantial: agents consume three to twenty times the AI Units compared to pure copilot deployments, based on observations from practice (source: SAP Licensing Experts). This distinction should be on the agenda in every AI Unit sizing discussion.
What are Foundation Models in the SAP GenAI Hub, and how do I select the right one?
The Generative AI Hub provides access to Foundation Models from multiple providers, including SAP-proprietary models such as SAP-ABAP-1 and SAP-RPT-1 as well as models from major external providers. SAP abstracts the model API, so no direct contract with the model vendor is required. The right model depends on the use case: SAP-proprietary models are optimized for SAP-specific tasks and carry lower token rates. External models provide more flexibility for generic tasks (source: SAP Help Portal, Generative AI Hub).
What does MCP mean for SAP contracts?
The Model Context Protocol is an open standard that can connect SAP agents and external AI toolchains. SAP has supported MCP since Q1 2026. Contractually relevant: the SAP API Policy v4/2026 does not permit autonomous third-party AI agents at SAP APIs. Third-party components may provide Skills but may not act independently. Organizations planning hybrid architectures should have this boundary clarified in the Order Form.
Is the free Joule Studio access through end of 2026 available for all customers?
SAP provides free design-time access to Joule Studio through end of 2026 under fair-use conditions. This covers the development environment, not production operations. Runtime licenses for customer-built agents running in production must be procured separately. Organizations building agents in Joule Studio during 2026 should factor runtime costs for 2027 into budget planning early.
Next Steps
How the Joule components are licensed in your SAP contract, which AI Unit allowances are available for Skills and Agents, and where governance moments are not yet reflected: this can be assessed systematically.
Schedule a Contract Check or use FinOptory AI for an initial assessment.
Further articles in this pillar:
Next Steps
If you would like your current contract reviewed for risks and available commercial levers: the FinOptory Contract Check is a fixed-price engagement that delivers a structured basis within four weeks.
