Agentic AI has a price: it's called ERP migration

Every few years, enterprise software discovers a future it insists you cannot afford to miss.

Often these claims outrun the product, and amid the hype, organizations commit time and money to innovations that never quite live up to their billing.

Agentic AI is the exception, and that is exactly why it deserves a closer look.

We're not talking about chatbots that draft emails or summarize reports. This is something entirely different: software that acts.

Agentic AI can read a company's own approval hierarchies and permissions, make decisions within finance or supply chain processes, and execute them.

The last wave of enterprise AI offered suggestions. This one does the work. For the organizations that get it right, that is a different order of advantage.

The large vendors know this, which is why they have spent the past year racing to own agentic AI. At its Sapphire conference this spring, SAP unveiled what it calls the autonomous enterprise: more than 200 specialized agents that carry out tasks across the core business functions, orchestrated by some 50 domain-specific assistants and reached through a single interface, Joule.

Oracle has been building the same capability into Fusion, its cloud suite for finance , HR and supply chain.

Unlocking the potential

There has been no shortage of talk about what that could unlock, and for many organizations the excitement is well earned: an agent that can carry out work, not just recommend it, changes what the software is for. But for most of them, already running this software, capability was never going to be the sticking point. Access is.

As the major vendors have built it, agentic AI is native to the system of record, woven into the core platform that runs the business. That is a real achievement: an agent that respects your permissions and executes a live transaction is worth far more than one bolted from outside.

But "native" also carries a second meaning – and this one never makes the keynote speech. With the agent tethered to the vendor's cloud platform, there can be no reaching it without committing to that same platform, regardless of what your systems run on today.

SAP shows how that gate works. Until this spring, Joule reached only customers on its cloud subscriptions: RISE and GROW. At Sapphire, with much of its installed base showing little sign of moving, SAP opened a door, but a narrow one. ECC and S/4HANA on-prem customers are no longer shut out, provided they commit to moving the majority of their SAP estate to Cloud ERP .

Even then, they get only a limited set of the AI capabilities rather than the full portfolio. The on-prem route, in other words, is sold on the condition that you start paying for the cloud. And for the older ECC core, a clock is running regardless: mainstream maintenance ends in December 2027, with a stay of execution to 2030 at a premium, after which the choice narrows to an unsupported system or an upgrade.

So an ECC customer that takes the on-prem route is still paying for the cloud to access AI capabilities, even as support drains out of the system that they rely on (according to a timeline they never chose).

A more direct route

Oracle arrives at the same place by a more direct route: its agents do not exist outside the cloud. The Fusion agentic applications it has rolled out this year run only inside Fusion Cloud, on Oracle's own infrastructure and within its security model. There is no on-prem edition to license. An organization still running E-Business Suite cannot switch these agents on where its systems sit today; to use them at all, that organization will have to re-platform onto Fusion.

Strip away the packaging and the structure is the same in both cases: the on-ramp to agentic AI is the migration these vendors have been trying to sell all this time. What is new is the leverage. Cost, risk and disruption have held many organizations on-prem for years – despite all the pressure to move. Could the prospect of being shut out of agentic AI be the argument that finally overcomes their objections?

The budget is already committed

The migration and the AI draw on the same budget, and the migration has first claim on it. In the Americas' SAP Users' Group, 61% of members reported that budget was the biggest challenge they faced this year, and the group's research director was blunt about the cause: the cloud ERP projects are themselves creating the pressure, with AI expected to land on the same budgets next.

The sequence is unforgiving. Pay to reach the platform, then pay again to use the AI once you are there, because the headline subscription covers only a limited band of embedded features, and the rest is metered by consumption. The capital a CIO would want to invest in building an advantage is spoken twice before a single agent has delivered a measurable outcome.

And that is before the program meets its harder test. Boards have grown tired of pilots that never reach production and spending that generates activity no one can tie to an outcome. In some cases, these will be genuine execution failures, but a program that starts short of capital, on a timetable set by someone else, is not starting from a position of control.

Who decides the order of operations?

A CIO should separate the two decisions the vendor has deliberately combined. Whether to modernize is one question. When to do it, in what order, and against which budget is another. Nothing requires that the second be dictated by an end-of-support date printed on someone else's roadmap.

The critical issue here is who holds the authority to decide how the organization's most critical systems change and when. When a vendor sets the timetable, the sequence and the price of innovation, the executive accountable for that estate is not really the one running it anymore. Restoring that authority does not mean refusing to modernize. It means refusing to let the vendor selling the upgrade also decide when and how you buy it.

This is th...

Indexed and credited by AIPROPX. Originating outlet: TechRadar. Open at source →