Can AI summarise this? Can it create that? Can it answer a question faster? Can it help an employee complete a task? Those questions still matter. But they are no longer the most important ones.
As AI moves deeper into business operations, leaders need to start asking something more significant: which decisions and actions should technology be trusted to carry out and under what conditions?
That is where the idea of the Autonomous Enterprise becomes relevant. SAP’s vision goes beyond placing AI alongside existing applications. It brings together Joule, AI assistants and agents, connected applications, business data and governance so that defined parts of the organisation can sense what is happening, understand the business context and act.
It sounds like a technology story. In reality, it is much more of an operating model story.
Autonomy Does Not Mean An Enterprise Running Without People
The word ‘autonomous’ can create the wrong impression. It can sound like a future where AI runs the company while people watch from the sidelines. That is neither a particularly useful interpretation nor, for most organisations, a desirable one.
A more practical way to think about autonomy is to look at how work moves through a business today.
An event occurs. Someone notices it. Information is gathered. A decision is made. Another person is contacted. An approval is requested. A transaction takes place. The process continues.
Sometimes all of that happens quickly. Sometimes it involves three systems, several emails and the familiar question: ‘Who owns this now?’
The Autonomous Enterprise changes that operating rhythm.
Applications identify signals as they occur. Business data provides the context required to understand what those signals mean. Assistants coordinate activity and specialist agents carry out defined actions. People remain responsible for setting direction, approving significant decisions and handling the situations where judgement matters.
That distinction matters. The objective is not maximum autonomy. It is appropriate autonomy.
The Real Change Is From AI Assistance To Business Execution
Many organisations have already introduced AI into individual tasks. An employee asks a question. AI provides an answer. A user requests a summary. AI produces one. A system identifies an anomaly. Someone decides what happens next. These applications can be valuable but the person still sits at the centre of each action.
Agentic AI changes the model because technology can move beyond identifying or recommending an action and begin carrying out defined work across a process.
Consider a supply chain example. A connected application identifies a shipment delay. Rather than simply flagging the issue to a planner, the wider business context can be considered: available inventory, customer commitments, alternative supply and the impact on production.
An assistant can then coordinate the relevant agents to initiate appropriate actions within predefined rules. A person becomes involved when an exception falls outside those boundaries or when the commercial implications require human judgement. The same principle can apply across finance, procurement, HR and customer operations.
The important part is not that an agent completed a task. It is that the organisation has designed a process in which the right combination of applications, data, AI and human judgement can respond to an event coherently. That is a much larger shift.
Which Means Buying More AI Is Not The Starting Point
There is an understandable temptation to begin an Autonomous Enterprise discussion with the agents themselves. What agents are available? What can Joule do? Which capabilities are coming next? Those questions have their place but starting there risks treating autonomy as another technology deployment. Before deciding where an agent should act, organisations need to understand the process it will be acting within.
Question #1 Where are decisions currently made?
Question #2 Which activities are repetitive and rules based?
Question #3 Where do delays occur because somebody has to identify an issue and move work forward manually?
Question #4 Which decisions require judgement or accountability that should remain with a person?
Question #5 What data is needed before an action can be taken with confidence?
Question #6 Where does the process cross systems, teams or organisational boundaries?
Question #7 Those questions are less exciting than a demonstration of a new AI capability. They are also far more likely to determine whether that capability creates value.
Governance Moves From Policy To Process Design
Governance is often discussed as something surrounding AI. Policies are written. Committees are established. Usage is monitored. In an Autonomous Enterprise, governance also needs to exist within the process itself.
An organisation needs to decide what an agent is permitted to do, which information it can use, when approval is required and what happens when an exception occurs. That creates practical questions.
Question #1 Should an agent recommend a purchasing action or place the order?
Question #2 Can it resolve an invoice discrepancy below an agreed value without intervention?
Question #3 Can it change a supply plan when demand moves outside an expected range?
Question #4 At what point does an action need human approval?
There is unlikely to be one enterprise wide answer. Different processes carry different financial, operational, regulatory and reputational consequences. The level of autonomy should reflect that.
For some activities, full execution within clearly defined rules may make sense. For others, the appropriate role for AI may remain recommendation and support.
That is why success should not be measured by how many processes become autonomous. It should be measured by whether each process has the right level of autonomy for the outcome and risk involved.
People Will Still Make The Decisions That Matter
There is another reason to frame this as an operating model change. Roles will change. If agents increasingly handle routine coordination and transactional work, people spend less time moving information between systems and more time dealing with exceptions, interpreting outcomes and making decisions where judgement matters.
In finance, that could mean less effort coordinating routine close activity and more capacity to investigate exceptions or advise the business.
In supply chain, it could mean planners spending less time identifying disruption and more time making decisions where customer, supplier or commercial trade-offs are involved.
In procurement, routine interactions and approvals may increasingly be handled within agreed policies while professionals concentrate on supplier strategy, negotiation and risk.
The technology is only one part of that change. Organisations need to consider responsibilities, controls, skills and measures of performance. Automating part of a process without redesigning the work around it can simply move the bottleneck somewhere else.
Start With A Business Challenge, Not An Ambition To Become Autonomous
No organisation needs to wake up one morning and declare that it will become an Autonomous Enterprise. The transition is far more likely to happen process by process and domain by domain.
SAP’s own direction reflects this through autonomous domains across finance, spend, supply chain, human capital management and customer experience. The internal model also considers domain blueprints as a way to translate the wider Autonomous Enterprise vision into practical capabilities for a defined area of the business.
For business leaders, a sensible starting point is therefore quite focused. Choose an operational challenge where there is a clear outcome to address. Understand the process behind it. Identify the decisions, signals and actions involved. Assess whether the data provides enough context. Define what AI should be permitted to do and where people need to remain involved. Then determine whether assistants and agents can change the economics or performance of that process. This is a much stronger foundation than starting with a list of AI features and searching for somewhere to use them.
The Organisations That Progress Will Treat Autonomy As A Capability
There will be plenty of discussion about autonomous technology over the coming years. New agents will appear. Existing capabilities will mature and the boundaries of what can be carried out by AI will continue to move.
The more important question for organisations is whether their operating model is ready to make sensible use of that capability. Technology can identify signals. Data can provide context. Assistants can coordinate activity and agents can carry out work but somebody still needs to decide what good looks like. Somebody needs to define the rules. Somebody needs to determine which outcomes matter and somebody needs to know when human judgement should take over.
That is why the Autonomous Enterprise should not be viewed as a destination where every business process eventually runs itself. It is a progression towards organisations that can decide, deliberately and confidently, where autonomy creates value and where people remain essential. The companies that approach it this way are likely to have a far more productive AI conversation.
Not simply: what can the technology do?
But: what should we trust it to do for the business?
Where Should Your Organisation Begin?
Moving towards more autonomous operations requires a clear view of your processes, data, technology landscape and governance.
Birchman helps organisations identify where SAP AI, Joule, assistants and agents can address real operational challenges and where the foundations need attention first.
Talk to our team about your SAP Autonomous Enterprise priorities. We are here when you are ready.
