SAP Connect 2026 in Las Vegas offered a pretty clear look at where SAP believes enterprise AI is heading next: toward more autonomous operations, with agents working across finance, procurement, supply chain, HR, customer experience, and other core business functions.
But getting there raises a harder question.
AI agents are getting better fast.
Yet if they are going to start touching critical parts of the business, “better” is not really the standard.
It has to be right. And it has to be right consistently, too.
That tension ran through much of the keynote.
SAP spent the session showing what it calls the Autonomous Enterprise, with agents working across key business functions.
And there was a clear thread running through all of it: how do you give agents enough context to make the right decisions, and enough control to trust them with more autonomy?
Christian Klein, SAP CEO, framed the first part early in the keynote. Frontier models are getting better, but they still “don’t know how your business works.” Ask one which of your suppliers will deliver next week and, without that business context, it will guess.
Which raises the first question:
If LLMs are probabilistic, how do you get the right answer every time?
Then SAP Chief Security Officer Marielle Ehrmann brought up the other side of the problem. As agents move from answering questions to taking actions inside critical processes, reliability is not enough. You also need control.
She summed it up with another question:
“How fast can we deploy AI without making tomorrow’s headlines?”
Those two questions capture much of the logic behind what SAP showed at Connect: give agents the business context they need to make reliable decisions, then add the governance, traceability, and human oversight required to safely increase autonomy.
Here’s a closer look at the main takeaways from SAP Connect 2026.
A Good Answer Is Not Enough
One of SAP’s clearest messages was that enterprise AI cannot depend on the frontier models alone.
LLMs are probabilistic by nature. They can reason, interpret language, generate recommendations, and produce answers that sound convincing.
But sounding right and being right are not the same thing. SAP made that distinction explicit during the keynote: for many enterprise tasks, “a good answer is not enough.”
If an agent is involved in moving money, making a procurement decision, or taking action inside an operational process, the result also needs to be correct, compliant, and traceable.
SAP is not trying to make the LLM deterministic
Giving an LLM more context does not suddenly make the model deterministic.
LLMs still generate outputs probabilistically, predicting the next token based on the information available to them.
SAP’s approach is to reduce the uncertainty around those outputs by grounding agents in the business context they need to run.
That means giving them access to the same data, business rules, permissions, processes, and relationships that already define how the company works.
So when an agent moves into a real business process, it is not free to “invent” how that process should run.
Business Context Is Where SAP Thinks It Has an Advantage
This explains why context came up so often during SAP Connect.
A generic model may know what a purchase order is.
It does not automatically know your suppliers, approval thresholds, inventory position, company policies, permissions, contracts, master data, or how one business event connects to another.
That business context is what SAP wants its agents to understand.
Business AI Platform: Build, Contextualize, Govern
SAP positioned the Business AI Platform as the layer where companies can build, contextualize, and govern AI across their landscape.
Joule Studio handles the build side, allowing companies to extend SAP agents or create their own.
Then Business Data Cloud and Knowledge Graph provide the context.
SAP described this layer as a way for agents to understand business data, processes, policies, relationships, and logic across SAP and third-party environments.
Knowledge Graph, in particular, was presented as the map that helps agents navigate processes, data flows, business objects, relationships, and rules.
Instead of explaining the company from scratch every time, the goal is for the agent to work with a structured understanding of how the business actually works.
Joule Work Becomes the AI Engagement Layer
That business context comes together in Joule Work, which SAP presented as a new AI engagement layer across the enterprise.
Christian Klein summed it up with a simple phrase: “a new UI for AI.”
The idea is pretty straightforward.
For years, users had to learn how SAP worked: transaction codes, exports to spreadsheets, presentations, and a lot of manual steps in between.
With Joule Work, SAP wants that interaction to feel much more natural.
You can talk to SAP in your own language, ask questions, delegate tasks, analyze SAP and non-SAP information, and let Joule connect you with the right assistant or agent.
SAP organized the experience around three areas: Conversations, for asking questions and delegating work; Business Insights, for analyzing and visualizing business information; and Joule Studio, where companies can extend existing SAP agents or build their own.
The important part is that Joule is not just another chat interface sitting on top of the system.
It is the place where users can interact with the data, business context, agents, and workflows already running underneath.
110,000 SAP employees are already using Joule
SAP also brought its own internal usage numbers to the stage.
According to the keynote, around 110,000 SAP employees are already using Joule to perform hundreds of thousands of tasks every day.
SAP reported productivity gains of more than 20% across areas including HR, finance, procurement, and supply chain.
Klein used earnings preparation as one example.
What used to take weeks of coordination across finance, investor relations, and communications can now be supported by Joule using internal financial information and external analyst data, surfacing relevant KPIs and helping anticipate questions.
The larger point is that SAP does not want Joule to be just another chatbot.
It wants it to become the layer through which users increasingly delegate work across the business.
From Individual Agents to an Autonomous Suite
SAP also made clear that the conversation is moving beyond individual agents.
The company said it has more than 20 assistants and over 200 agents generally available, with a target of more than 400 agents by the end of the year.
Those assistants and agents are being organized across major business domains including finance, spend, HCM, supply chain, customer experience, and industry AI.
But the more important idea is orchestration.
A single agent does not need to know how to do everything.
The assistant can identify the task, select the right specialized agent, connect it to the relevant context, and move the work through the right business process.
That is a very different model from simply adding a chatbot on top of an application.
Procurement: More Analysis, Less Manual Coordination
Procurement offered one of the clearest examples of how SAP sees agents being applied in practice.
A customer described the challenge of managing hundreds of categories and suppliers while creating RFPs, analyzing bids, and preparing negotiations.
At scale, those processes take time and can become inconsistent, particularly in highly regulated environments where traceability and auditability matter.
The company has been piloting SAP’s sourcing assistant around bid analysis and negotiation support.
The goal is for agents to handle more of the analytical heavy lifting: creating bids, comparing offers, identifying negotiation levers, and bringing more consistency to work that previously required significant manual effort and coordination.
The customer said it expects a 25% reduction in cycle times and a 35% productivity uplift.
But the advice around deployment was just as important as the expected gains.
- Start with reliable data.
- Choose processes where a wrong decision is recoverable.
- Involve the people who know where the actual friction is.
- And do not underestimate change management.
That message came back several times during the keynote.
Automation does not compensate for weak foundations.
Supply Chain: From Reactive to Predictive
Morgan Foods gave one of the strongest examples of what those foundations can enable.
Before its SAP modernization, much of its supply chain planning was manual, siloed, and dependent on Excel.
With S/4HANA and IBP, the company connected more of the operation across forecasting, ingredients, production scheduling, warehousing, distribution, and supplier collaboration.
From 3–4 weeks to about 5 minutes
Morgan Foods, one of the companies invited to share their case at the stage, said it once took three or four weeks to determine whether it could fulfill a large customer request.
Today, that can be answered in about five minutes.
The important part is how they got there.
This was not the result of simply putting an agent on top of fragmented systems.
The company first built a more connected operational foundation.
Now that same foundation can support the next shift.
Morgan Foods CFO Steve Henke described it as moving from reactive to predictive.
Agents can connect demand signals, inventory levels, and supplier status much faster than a person or team could.
But Henke was also direct about the prerequisites: the right data, good master data, and clean, documented business processes.
And people still matter.
Henke emphasized that humans need to remain above the agents because AI cannot replace judgment and experience.
Autonomy Starts Connecting Functions
This is where SAP’s Autonomous Enterprise narrative becomes more interesting.
The goal is not just to automate departments individually.
SAP wants an event in one part of the business to trigger coordinated actions across several others.
One keynote scenario used an acquisition as the example.
Bringing new employees into the company was not treated as an isolated HR task.
It also affected IT, facilities, procurement, access, devices, and supplier availability.
When laptops were at risk of arriving late, the workflow could identify the problem, look for an alternative supplier, route the necessary approval, and continue the onboarding process.
That is much closer to how real business processes behave.
They cross functional boundaries.
And that is exactly where context becomes more important.
TechWolf Adds More Workforce Context
SAP also announced the acquisition of TechWolf, expanding its capabilities around skills and workforce intelligence.
The idea is to improve visibility into which skills already exist inside an organization, where gaps remain, how roles are changing, and where people could develop or move next.
SAP said it expects TechWolf intelligence to be integrated into SuccessFactors across workforce planning, learning, mobility, recruiting, and the broader SAP ecosystem.
Again, the pattern is consistent.
More context gives the system more information about what already exists before another decision gets made.
Before hiring externally, leaders can better understand internal capabilities.
Once someone joins, the same context can help identify development paths, learning opportunities, and future roles.
SAP Pay Brings Agents Closer to Execution
Finance also brought one of the event’s most concrete announcements.
Christian Klein introduced SAP Pay, a payment service embedded directly into SAP Cloud ERP.
He framed the problem in familiar terms.
Paying a supplier can mean approving an invoice in one system, initiating a bank transfer somewhere else, and then reconciling everything afterwards.
Three separate steps for something that, in Klein’s words, should simply happen.
SAP Pay is designed to bring those steps into the same flow.
Once an invoice is approved, SAP agents can execute the payment and reconcile it automatically inside the ERP. SAP also said the service will support several payment methods, including cross-border payments and digital currencies.
The company also pointed to the potential for around a 25% TCO reduction on average.
This matters because agents are moving beyond recommendations.
They are getting closer to execution.
And that makes the second question from the keynote much more important.
How Fast Can We Deploy AI Without Making Tomorrow’s Headlines?
SAP Chief Security Officer Marielle Ehrmann put the issue directly.
A year ago, she said, boardrooms were asking:
How fast can we deploy AI?
Now the question has shifted:
“How fast can we deploy AI without making tomorrow’s headlines?”
That is the other side of autonomy.
Once an agent can take action in a process, access business data, interact with systems, or execute a transaction, you need much more than a capable model.
- You need to know who owns it.
- What it can access.
- What actions it is allowed to take.
- What level of risk it carries.
- And who can step in if something goes wrong.
SAP Calls It “Freedom Within Boundaries”
Eman described SAP’s approach as “freedom within boundaries.”
Agents operate within identity and access controls.
They inherit the authorizations of the person they represent and can only access data and perform actions they are explicitly permitted to execute.
SAP also emphasized governance and visibility.
As companies deploy hundreds of agents across SAP and non-SAP environments, they need to know who owns each one, what it can access, and what risk it carries.
In this model, governance is not a passive compliance checklist added at the end.
It becomes part of how the agent operates.
What Does It Take to Get There?
For all the talk about agents and autonomy, SAP’s answer came back to some fairly traditional foundations.
- You need good data.
- You need good processes.
- And you need people to change how they work.
1. Get the Data Right
If the underlying data is fragmented or unreliable, giving an agent access to more of it does not solve the problem.
Several speakers returned to trusted data, master data, harmonization, and unified business context.
One customer explained that when data is fragmented across legacy systems, there is very little for AI to work with.
AI starts becoming useful at scale when the underlying data is harmonized, trustworthy, connected, and prepared for it.
2. Get the Processes Right
Agents also need clarity around how work is supposed to happen.
That means documented processes, defined approval paths, business rules, and a clear understanding of when people need to intervene.
Morgan Foods put it simply: technology delivers when the foundation is there, including the right data, master data, and clean business processes.
3. Do Not Underestimate Change Management
And then there is the human side.
Christian Klein returned to this near the end of the keynote.
AI, he said, requires significant change management if companies want it to deliver the expected outcomes.
Based on his own experience driving transformation inside SAP, he described change management as one of the hardest parts of the job.
That may be one of the easiest points to overlook.
You can have a technically strong agent and still fail if people do not trust it, understand how to use it, know where they remain accountable, or change the surrounding process.
The Road From Semi-Autonomous to Autonomous
Two questions probably summarize the keynote better than anything else:
- Can an agent give you the right answer consistently?
- Can you trust it enough to give it more autonomy?
Those questions matter because CIOs around the world are trying to answer them inside their own environments.
And as SAP made clear throughout the keynote, the answer starts with the foundations.
The data has to be ready. The processes have to be clean and optimized. And the business context needs to be structured in a way that agents can actually use.
Then comes change management.
Teams will need to adapt to new ways of working, where people increasingly supervise, review, and guide the process while agents take on more of the execution.
That is where many companies are today: in a semi-autonomous stage.
SAP’s vision is that agents will gradually take on more complex work, moving from assistance to execution and helping compress processes that once took weeks or months into days, minutes, or even seconds.
But that level of autonomy only works if the foundations are ready.
At Inclusion Cloud, we have been working with SAP for more than 20 years. If you are looking to accelerate your agent implementation with the right business context, we can help with certified SAP, data, and AI resources. Book a call with us!