← All posts

Joule, Assistants, Agents and Skills: Who Does What

By Srinivas Vanamala · 16 November 2026 · 7 min read

In every SAP conversation about AI, four words come up: Joule, assistant, agent and skill. Many people use them as if they mean the same thing.

So consultants ask:

When someone says “SAP has an agent for that”, what is actually doing the work?

The short answer: there are four levels. Joule is where the person states the request. A Joule Assistant understands the intent and coordinates. Joule Agents plan and complete multi-step tasks. Joule Skills do the single, fixed operations, usually through an API.

Let’s understand this through a simple example.

1. The big picture

Think about a buyer in purchasing who wants to know why an invoice from a supplier is still not paid.

Today, the buyer has to:

  • Open the purchase order.
  • Check the goods receipt.
  • Look at the invoice and its status.
  • Find out why it is blocked.
  • Ask accounts payable to release it.
  • Tell the supplier.

Six steps, several transactions, and at least two teams.

Now imagine the buyer simply asks the question in plain language, and the system does the looking up, works out the cause, and proposes the fix.

That only works if the four levels each do their own job. One understands, one coordinates, one reasons, and one executes.

2. Who does what

Let’s separate the responsibilities.

Joule is the place where people say what they need, in their own words.

A Joule Assistant belongs to a domain or role. It interprets the intent and coordinates the agents that are needed.

A Joule Agent is a specialised worker. It plans, acts and completes a multi-step task across systems.

A Joule Skill is a small, predefined operation. It does one thing the same way every time.

SAP’s learning material draws the line clearly. A skill does one predefined task in a deterministic, predictable way. For more complex, goal-driven requests, Joule calls on an agent, which reasons through the problem, makes a plan and runs the right skills in sequence.

Illustrative hierarchy: from request to system
Business userStates what is needed
↓
JouleReceives the request in plain language
↓
Joule AssistantUnderstands intent • Coordinates agents for one domain
Joule AgentPlans and completes a task
Joule AgentPlans and completes a task
↓
Joule SkillsOne fixed operation each, through an API
↓↑
Business systemsSAP S/4HANA • other SAP applications • non-SAP systems. They hold and process the business transactions.
Conceptual illustration only. SAP's product architecture contains additional services and layers.

3. Let’s see a practical example

The buyer asks:

“Why is the invoice from supplier 100245 not paid yet?”

In a possible AI-enabled process, this is what happens.

Step 1: Joule receives the request. The buyer types or speaks the question.

Step 2: An assistant takes over. An assistant for the spend domain recognises that this is about an invoice and decides which agent should work on it.

Step 3: An agent plans the steps. The agent decides it needs the purchase order, the goods receipt and the invoice status.

Step 4: Skills fetch the facts. One skill reads the purchase order. Another reads the invoice. Each skill calls an API, for example an OData service in SAP S/4HANA.

Step 5: The agent works out the cause. It compares the documents and finds, for example, a quantity difference between goods receipt and invoice.

Step 6: A person decides. The agent proposes the correction. Releasing a payment block changes data, so a person with the right role confirms it.

This is an illustration, not a description of one specific SAP agent. What runs in your system depends on your products, integrations, configuration and authorisations.

How is this different from classic automation? Here is an illustrative comparison. A classic workflow follows a fixed path: if the invoice is blocked, send an email to accounts payable. It does not look for the reason. The agent in our example chooses which documents to read and compares them before it proposes anything. If the cause had been a missing goods receipt instead of a quantity difference, the same question would lead to a different plan.

4. Which pieces matter

PieceRole in our example
JouleReceives the buyer’s question
Joule AssistantChooses and coordinates the right agent
Joule AgentPlans the steps and finds the cause
Joule SkillsRead the purchase order and the invoice
Actions and APIsHow each skill reaches the backend
Joule Studio in SAP BuildWhere custom skills and agents are built

These pieces do not do the same job. The agent decides. The skill executes.

SAP’s learning material gives a real example of scale: the Financial Closing Assistant orchestrates six agents, and the Recruiting Assistant coordinates five.

5. What can go wrong?

A skill is chosen largely by its description. SAP says Joule matches a request to a skill using the skill’s description, the user’s role, the conversation and the business context.

If two skills have vague or similar descriptions, the wrong one can be chosen. If there is no API behind a skill, it has nothing to call. And because an agent acts with the user’s own permissions, the solution should be designed to stop and explain clearly when a role is missing.

So the weak points are rarely the AI model. They are unclear descriptions, missing APIs and missing roles.

6. What does this mean for SAP consultants?

A functional consultant decides which tasks are simple and rule-based, and which need reasoning. That decides skill or agent.

An ABAP or integration consultant makes sure the API behind each skill exists, is released and is secured.

A BTP developer builds the skills and agents in Joule Studio and writes the descriptions that make them findable.

A project lead or architect asks where a person must confirm. SAP describes agents that work within the user’s authorisations, log every decision for audit, and ask for human approval at key decision points, and says the customer decides the level of autonomy.

The important skill is breaking a business task into small, clear operations that a machine can choose correctly.

7. Where should you start? Design one skill on paper

Pick one small task from your own work and fill in this card.

FieldWorked exampleYour answer
Task in one sentenceShow the open purchase orders for one supplier.
When should Joule use it?When a buyer asks which purchase orders are still open for a named supplier. (One specific sentence. If you need the word “anything”, split the task.)
InputsSupplier number.
Which API gives this data?A purchase order API for SAP S/4HANA, found in SAP Business Accelerator Hub.
Read only, or does it change data?Read only, so no one needs to confirm. If it changed data, a person would confirm first.
What comes back to the person?A list: purchase order number, date and open value.

If you can fill in every row, a developer can build it. If you cannot name the API, you have found a real gap early.

What is live today

In May 2026, SAP outlined more than 50 Joule Assistants and over 200 agents. In late October 2026, at SAP TechEd, SAP announced 40 assistants and 300 agents between SAP and its partners. The numbers differ because the announcements differ in date and scope. SAP’s own learning material says availability spans general release, early access and planned releases across 2026.

Always ask whether a specific assistant or agent is available for your product and release.

Final thoughts

Joule, assistant, agent and skill are not four names for one thing.

They are four levels with one job each. The applications still hold the transactions, and people stay responsible.

The next time someone says “SAP has an agent for that”, ask which assistant coordinates it and which skills it calls.

Then you will know who is actually doing the work.

Sources from SAP (free to read)

Next step Learn SAP Autonomous Enterprise Step by Step Want the full picture? This Udemy video course explains the applications, data, AI, agents and architecture in a simple, beginner-friendly way. Explore the Course on Udemy → Ideal for SAP consultants who want the complete picture. Beginner-friendly lessons Clear explanation of apps, data, AI and agents Practical understanding for SAP consultants

Have Questions?

Get in touch with any questions about the courses or your learning journey.

Contact Srini