Guide9 min readPublished on 2026-07-08

What are Joule agents and how do they work

Guide to Joule Agents, SAP's AI agents: architecture, orchestration, A2A and MCP protocols, and how they integrate with S/4HANA, SuccessFactors and Ariba.

In a nutshell

Joule Agents are SAP's specialized AI agents, each dedicated to a specific business function (finance, supply chain, service management). Joule acts as orchestrator: it decomposes tasks, selects the right agent and coordinates the workflow. Agents communicate with each other via the Agent2Agent (A2A) protocol and access SAP systems through the Model Context Protocol (MCP). They integrate with S/4HANA, SuccessFactors, Business Data Cloud, Concur and Business Network, and can also be reached from external platforms (Google Cloud, Azure, AWS, IBM Cloud) via Agent Gateway.

Not a single agent, but a network of specialists

The most common mistake is imagining Joule as a single do-everything assistant. The reality is different: SAP built a multi-agent architecture, where each Joule Agent specializes in a precise business function — finance, supply chain, service management, HR — instead of handing everything to one generalist agent.

The logic is the same as in a real company, where you don't ask the same person to close the books and handle a supplier return: different skills are needed. The advantage of this approach is that each agent can specialize deeply in its own domain, instead of being mediocre at everything.

Joule as orchestrator

Above the network of specialized agents, Joule acts as central orchestrator. SAP describes this role as a "conductor": it decomposes a complex task into sub-tasks, selects the most suitable agent for each one, and coordinates the overall workflow.

Humans remain in the "composer" role: they define the goal and constraints, while Joule orchestrates execution. It's a model that cleanly separates two responsibilities — what needs to be done (human) and how to coordinate who does it (Joule) — reducing manual oversight without removing control.

Want to know which Joule Agents your business needs?

30 minutes to discuss your specific case.

Book a call

The two protocols holding it all together: A2A and MCP

Two open standards make this architecture possible. The first is Agent2Agent (A2A), the protocol SAP adopted as its preferred standard for multi-agent collaboration and vendor-to-vendor interoperability: it lets one agent delegate a task, ask another agent what capabilities it has, and exchange information in a structured way, regardless of the framework it was built with.

The second is the Model Context Protocol (MCP), Anthropic's standard governing how an agent discovers, understands and interacts with external tools. SAP uses it internally to give Joule Agents semantically enriched access to SAP business capabilities — not just raw data, but context already structured around what that data means for the business process.

Where the agents integrate

Joule Agents integrate natively with SAP S/4HANA, SAP SuccessFactors, SAP Business Data Cloud, SAP Concur, SAP Customer Experience and SAP Business Network. This is the core of the SAP ecosystem, where the heaviest-weight processes live.

But the architecture doesn't stop at SAP's boundaries: external platforms like Google Cloud, Microsoft Azure, AWS and IBM Cloud can consume Joule Agents through a dedicated Agent Gateway, or have Joule orchestrate them directly. It's an architecture designed for a company with multiple systems, not just pure SAP.

Why this architecture matters for adoption decisions

Understanding this structure changes how you evaluate a Joule agent project. You're not buying "a smarter chatbot": you're adopting an architecture of specialized, orchestrated agents that speak a common protocol (A2A) and access your systems through another common protocol (MCP).

This means the real implementation work isn't "turning on Joule," but deciding which agents you need, on which processes, with what boundaries of autonomy. It's the question we address together with our clients in every SAP Joule agent project.

FT
Federico Thiella·Founder, Maverick AI

Works with European companies on Claude and Anthropic ecosystem adoption. Has led AI implementations in private equity, consulting, manufacturing and professional services.

LinkedIn

Want to know which Joule Agents your business needs?

We analyze your SAP processes and identify where a specialized agent generates real value. Discover AI agents on SAP Joule

Write to us

Frequently asked questions: Joule Agents

No. SAP built a multi-agent architecture: each Joule Agent specializes in a specific business function (finance, supply chain, HR, service management), and Joule coordinates the network as a central orchestrator.
It's the standard SAP adopted for collaboration between different agents and vendor interoperability. It lets one agent delegate tasks, ask about another agent's capabilities, and exchange information in a structured way, regardless of the framework it was built with.
MCP governs how agents discover and use external tools and data. SAP uses it internally to give Joule Agents semantically enriched access to SAP business capabilities — not just raw data but structured context.
Natively with S/4HANA, SuccessFactors, Business Data Cloud, Concur, Customer Experience and Business Network. They can also be reached from external platforms like Google Cloud, Azure and AWS through an Agent Gateway.
Not the activation of the tool itself, but which specialized agents you actually need, on which business processes, and with what boundaries of autonomy and control. That's the assessment work we do before any Joule agent project.

Stay informed on AI for business

Get updates on Claude AI, business use cases and implementation strategies. No spam, just useful content.

Want to learn more?

Contact us to find out how we can help your company with tailored AI solutions.

Anthropic implementation partner in Italy. We work with companies in PE, pharma, fashion, manufacturing and consulting.

Book an introductory call
SAP Joule Agents: what they are and how they work (2026) | Maverick AI