In this article
A new operating layer for the enterprise
After listening to the SAP Connect keynote today, one message stood out very clearly for me: SAP is no longer positioning AI as another capability that sits on top of enterprise applications. It is positioning AI as a new operating layer for the enterprise.
That is an important distinction. For the last couple of years, much of the enterprise AI conversation has revolved around copilots, prompts, productivity and individual AI use cases.
What SAP demonstrated today was a much broader architecture:
Business Applications + Business Data + Business Context + AI Agents + Governance + Human Judgment.
And that, in my view, is where the conversation around the Autonomous Enterprise starts becoming much more interesting.
The perspective in brief
- Enterprise AI is moving from answering questions to completing business processes.
- Trusted data and business context give agents a foundation to act.
- Governance, deterministic execution and human judgment make autonomy practical.
From answers to outcomes with Joule Work
The introduction of Joule Work represents a significant change in how SAP sees the future user experience. Instead of users navigating multiple applications, reports, transactions and workflows to accomplish something, the idea is increasingly:
Tell the system what business outcome you are trying to achieve — and let the system orchestrate the work.
- Conversations allow users to ask questions and delegate work.
- Workspaces bring together SAP and non-SAP information, analytics and collaboration.
- And Joule Studio provides the ability to extend existing agents or create new ones based on specific business requirements.
What interested me most was not simply the conversational interface. It was the shift from:
“Give me an answer” → “Help me complete the business process.”
That is a fundamentally different expectation from enterprise AI.
Business context is the real differentiator
One of the strongest themes throughout the keynote was context. A generic LLM can reason over information. But an enterprise agent needs to understand much more:
- What does this customer mean?
- What business process are we executing?
- Which data is authoritative?
- What is the relationship between these objects?
- What policies apply?
- What is this user authorized to do?
- What happened before this step?
- And what downstream process will be affected by the decision?
SAP’s argument is that its Knowledge Graph, Business Data Cloud, application semantics, APIs, process knowledge and domain models collectively provide that context. I think this is an extremely important point.
The future enterprise AI battle may not ultimately be about who has the largest model. It may be about who can provide the model with the best business context.
And this is exactly why I believe SAP Business Data Cloud becomes strategically important to the Business AI story.
AI without trusted data is experimentation. AI with trusted data but without business semantics is analytics.
AI with trusted data, semantics, processes, authorization and the ability to execute becomes something very different. It becomes operational.
Build. Context. Govern.
The SAP Business AI Platform story was presented around three major pillars:
BUILD
Joule Studio and the ability to build and extend agents from no-code through pro-code, while accessing multiple AI models.
CONTEXT
SAP Business Data Cloud, Knowledge Graph, domain models, master data and business process semantics providing the grounding agents require.
GOVERN
identity, authorization, compliance, auditability and AI Agent Hub providing governance across SAP and non-SAP agents.
For customers, I believe this architecture matters more than simply asking, “How many agents does SAP have?” SAP already has hundreds of agents and assistants and intends to continue expanding that portfolio. But enterprises will inevitably create their own agents as well.
Microsoft will have agents. Google will have agents. ServiceNow, Salesforce and other platforms will have agents. Customers and partners will build agents. Specialized AI companies will build agents.
The bigger enterprise problem therefore becomes: How do all these agents discover each other, exchange context, operate within enterprise permissions and remain governed?
SAP’s support for open agent interoperability and the AI Agent Hub becomes particularly relevant in that world.
Intelligence that can act within business rules
Another message I took away from the keynote was the distinction between probabilistic AI and deterministic enterprise execution. For many enterprise processes, “probably correct” simply isn’t sufficient. Think about:
- releasing a payment,
- approving a supplier,
- changing payroll,
- posting a financial transaction,
- modifying a production plan,
- executing procurement,
- or changing an employee record.
These actions require authorization, controls and an audit trail. SAP’s approach is to embed agents into the deterministic workflows of the business applications, allowing the agent to reason while the underlying enterprise system continues to enforce the business rules.
That combination of:
probabilistic intelligence + deterministic enterprise execution
could become one of the most important architectural patterns for enterprise agentic AI.
What this looks like in the business
I appreciated that the keynote moved beyond architecture into actual business scenarios.
Novartis
Novartis discussed its pilot of the Joule Sourcing Assistant for bid analysis and negotiation support, with the objective of allowing sourcing professionals to spend less time processing information and more time on strategy and supplier relationships.
Morgan Foods
Morgan Foods demonstrated the importance of connecting forecasting, inventory, production and supplier information — and moving supply-chain decision-making from reactive toward predictive.
Nestlé
Nestlé showed why the digital core matters. Harmonized processes and trusted data across a global S/4HANA landscape provide the foundation for AI-driven finance scenarios such as intelligent prioritization of collections and recommended next-best actions.
PwC
PwC discussed industry-specific AI and how business context can improve areas such as billing and engagement management.
The common denominator across all of these wasn’t simply AI. It was:
Process + Data + Context + AI + Execution.
Autonomy needs human judgment
“Autonomous” should not mean removing people from every decision. SAP repeatedly emphasized different levels of autonomy and human oversight.
For lower-risk, repetitive activities, agents can increasingly operate autonomously. For higher-risk decisions, humans remain in the loop — approving, monitoring or overriding actions. I think this is critical.
The goal of enterprise AI should not be to eliminate human judgment. It should be to eliminate the manual effort that prevents people from applying their judgment where it actually matters.
Perhaps one of the most important parts of the keynote was the discussion around trust. As organizations move from deploying a handful of copilots to potentially hundreds of autonomous agents, new questions emerge:
- Who owns an agent?
- What data can it access?
- What actions can it execute?
- What risk classification does it have?
- Who approved it?
- What decisions has it made?
- Can an auditor reconstruct those decisions?
And, most importantly:
Who can stop it?
That is why capabilities such as SAP LeanIX AI Agent Hub, identity and authorization controls, auditability and AI governance are not secondary capabilities. They are prerequisites for enterprise-scale agentic AI.
From recommendations to execution
SAP also announced SAP Pay, embedding payments directly into SAP Cloud ERP. On the surface, that may appear separate from the AI story. I actually see it as part of the same architectural direction.
If enterprise systems are going to move from recommending actions to actually executing outcomes, the platform increasingly needs native mechanisms to complete those transactions. The future ERP experience therefore becomes less:
Record → Analyze → Recommend
and increasingly:
Sense → Reason → Decide → Execute → Monitor.
That is a very different ERP.
Start with the foundation, then build the agent
There was another theme across virtually every customer discussion today: Get the foundation right.
Trusted data. Clean master data. Connected processes. A modern digital core. Integration. Governance. And change management.
This resonates strongly with what we see in customer conversations every day. There is a temptation today to begin an AI program by asking: “Which agent should we build?”
I think the better questions are:
- Which business process are we trying to improve?
- Where is the friction today?
- What measurable outcome do we want?
- Is the underlying data trustworthy?
- Are the systems and processes connected?
- What actions should AI be permitted to take?
- Where must a human remain accountable?
Only then should we ask which agent to build.
For partners, I believe this creates a significant opportunity — but also requires us to evolve. The opportunity is no longer simply implementing another SAP product or building another isolated AI proof of concept. Customers will need help connecting the entire chain:
Clean Core → Integration → Business Data Cloud → Business AI → Agents → Process Transformation → Governance → Adoption.
This is where SAP BTP, Integration Suite, Business Data Cloud, Joule/Joule Studio, LeanIX, Signavio and WalkMe start becoming parts of the same transformation conversation rather than individual technologies.
The next wave of value is contextual
And this is also where industry knowledge becomes increasingly valuable. The most useful enterprise agents will not simply “know SAP.” They will understand how an oil & gas operator manages an AFE, how a government agency administers grants, how a manufacturer responds to a supply disruption, or how a procurement organization evaluates supplier risk.
Industry context + enterprise context + AI is where I believe the next wave of value will be created.
We have spent decades teaching people how to operate enterprise software. We may now be entering a period where enterprise software increasingly learns how to operate around people. That does not mean ERP disappears. Quite the opposite. The digital core becomes even more important because autonomous AI needs something trustworthy to reason over and something deterministic through which it can execute.
The UI may become conversational. The work may become agentic. The processes may become increasingly autonomous. But underneath it all, trusted data, integration, business rules, security, governance and human accountability remain essential.
That, for me, was the real message from SAP Connect today. The Autonomous Enterprise isn’t simply about having more AI agents.
It is about connecting intelligence to the processes that actually run the enterprise — and giving that intelligence enough context, control and trust to act.
And that is where things are getting very interesting.
Continue the conversation#SAPConnect#SAP#BusinessAI#Joule#JouleWork#AgenticAI#AutonomousEnterprise#SAPBTP#SAPBDC#BusinessDataCloud#SAPBusinessAI#CleanCore#IntegrationSuite#JouleStudio#EnterpriseAI#DigitalTransformation
Further reading
SAP Connect 2026: event overview ↗
SAP News: SAP puts the autonomous enterprise to work ↗
This article reflects the author’s keynote observations and perspective. Product names and trademarks belong to their respective owners.
