The AI debate around ERP is rarely about technology.
It is about who is allowed to intervene in how organisations work.
We like to pretend the choice is between AI inside the ERP or outside it, but that opposition distracts from what really matters.
AI outside the ERP has complete freedom to calculate, combine and predict, but no responsibility to change anything. ERP is where inventory, money and commitments come together and where decisions are taken that are not optional.
ERP vendors are increasingly opening their AI agents through APIs. From the outside, analyses can be created, predictions made and proposals generated before anyone decides. It looks like freedom, but everything remains bound to the platform’s data model, process logic and governance.
Fully independent data layers remain, for now, largely aspirational. Vendors are building their own data platforms as a gravitational field: all applications converge there, AI generates insights, but control over structure and governance remains firmly with the platform.
ERP vendors do this to strengthen their modular ecosystems. Within their data model, process logic and governance, applications work together optimally. The real intelligence lies in the relationship between data, processes and responsibility, and that relationship is never fully visible to external parties, let alone enforceable. It is not optional, and it is certainly not free.
As a result, power does not shift, work does. More thinking, calculation and preparation happens outside the ERP. Decisions and execution remain inside it. Not because this is technologically convenient, but because this is where ownership, control and consequences meet.
As long as AI operates outside that core, it can be no more than insight. It shows what could improve, but it does not change what actually happens.
Organisations that keep AI outside the ERP risk creating a new form of Shadow IT: a parallel reality of insights that never lead to action.
Once AI is allowed to act, insights immediately touch processes, responsibilities and choices that organisations would rather leave untouched. This increases the need for human oversight: organisations must actively monitor how these autonomous agents behave.
The reality of 2026 is not a migration of AI away from ERP. AI outside ERP can generate insights, but only inside ERP are decisions taken and executed.
Those who deploy AI without touching that core gain visibility.
Those who continue to use ERP without AI pay an increasing price for stagnation.
The question is not where AI belongs, but whether an organisation is willing to take responsibility for what AI makes visible.