AI in ERP systems is changing character, from answering questions to executing tasks. In ERP, it is not merely a technical evolution. It is a shift in risk profile that changes what good governance must cover.
AI in ERP systems is changing character.
It is no longer just about assistants helping a user write, search or summarise. The next step is giving AI a clearer role within the core business logic: to qualify, propose, trigger, follow up and in some cases act within defined boundaries.
That sounds like a technical development. In practice, it is also a shift in risk profile.
When the system begins to act, it changes not only what the ERP platform can do. It also changes which questions must be asked before it happens. This article is about that very shift, what it implies, why ERP is a special case and which three areas determine whether agentic AI becomes a controlled improvement or a new source of risk.
The step everyone misses: from assistant to agent
The first wave of generative AI in business systems has largely been about assistance.
- Summarise a customer meeting.
- Suggest phrasing.
- Generate a report.
- Explain a field.
- Find the next step.
Useful, absolutely. But the core logic has been the same: the user asks for help, AI responds.
What is now emerging is something else. AI is not just used to support an individual, but to drive tasks towards a goal within the framework of processes, data and rules. It can break down an assignment into sub-steps, work against live business data and return when human judgement is required.
That is where the question becomes more interesting.
Because when AI moves closer to execution itself, it is no longer enough to talk about functionality. You must also talk about control.
Why ERP cannot tolerate "AI on a hunch"
It is easy to talk about AI in general terms, as if the same logic applies everywhere. In ERP, it does not.
ERP is a special case because the margin for error is so much smaller. A mistake in a core business system rarely stays as an incorrect answer on a screen. It can impact accounting, delivery flows, compliance, payments, inventory, approval chains or customer relationships.
This means that agentic AI in ERP is not primarily a question of how much can be automated. It is a question of what is permitted to be automated, under what conditions and with what safeguards.
The difference is far from minor. In an Office environment, the consequence of a poor AI suggestion is usually just a poor document. In an ERP system, the consequence of an incorrect decision is something that shows up on the balance sheet, in delivery precision or in the audit trail.
That is also why good governance around agentic AI in ERP is not a by-product. It is the core prerequisite.
Governance is not an appendix, it is the deliverable itself
When AI begins to act in or around ERP, governance is not an appendix to the project. It becomes a core part of the delivery itself.
Here are three areas that determine whether agentic AI becomes a controlled improvement or a new source of risk.