AI Won't Disrupt the ERP. It Will Disrupt the People Around It.

Every vendor keynote this year tells the same story: AI is transforming the ERP. Agents that close the books, draft the journal entries, resolve the exceptions. Watch the demos and you'd think the transactional core of enterprise software is being reinvented in real time.

Look under the hood and something else is going on. The core is barely moving, and there are good reasons it won't move fast. The real movement is happening off-stage, in the rows of the keynote audience: among the integrators, the consultancies, and the internal teams whose economics were built on a scarcity that AI is preparing to dissolve. The ERP market is heading somewhere, but the product is the wrong place to watch. Watch the people.

What the consensus gets right

Let me steelman the keynote first, because the capabilities are real and dismissing them would be lazy.

SAP used Sapphire 2026 to announce an explicit pivot away from building feature-rich applications and toward agents that run business processes, under a new Business AI Platform. Its own release notes count more than 30 specialized agents and over 2,500 Joule skills, with Joule Studio, the build-your-own-agent tooling, reaching general availability in early 2026. In a telling reversal, SAP even agreed to bring a significant share of Joule agents to ECC and on-premise S/4HANA landscapes for customers committed to the cloud journey, something it had long refused. Oracle, for its part, launched Fusion Agentic Applications in March 2026: coordinated teams of agents embedded in the transactional system (a Ledger Agent, a Payables Agent, a Payment Agent), built to execute inside existing approval hierarchies and policies.

So no, this is not vaporware. The agents exist, they ship, and buyers want them: one industry survey this year found nearly four in ten enterprise buyers now expect generative AI to be delivered primarily through agents. The consensus is right that something big is happening.

It's just looking at the wrong layer for the disruption.

Why the core will resist

An ERP is not a productivity tool. It is a system of transactional integrity: deterministic postings, auditable trails, legal compliance in every jurisdiction where you operate. Its entire value is that the same input produces the same output, every time, with a paper trail a regulator can follow.

AI, as it stands, is probabilistic. And nobody — not the CFO, not the auditor, not the regulator — wants probabilistic accounting entries. The vendors know this perfectly well; SAP's own agent communications lean hard on the words "repeatable" and "auditable," which tells you exactly what objection they're answering. That's why the agents shipping today sit at the edges of the core: they retrieve, they draft, they recommend, they pre-fill, and a human or a deterministic rule validates. The transactional engine underneath changes slowly — not because vendors are behind, but because slowness is what the core is for.

So the honest answer to "when will AI transform the ERP" is: later than the keynote implies, and by design. Which raises the better question — if the product moves slowly, where does all this capability actually land?

The market that hasn't repriced yet

Here's what I observe from inside the market, as someone who sells ERP expertise for a living: the consulting market has not repriced. Day rates hold. Project staffing models hold. The knowledge premium, what clients pay for the scarcity of people who know how these systems behave, is intact.

It holds because the vendor AI tooling is not yet commoditized inside client IT departments. Joule and its equivalents are shipping, but they are not yet sitting under license on every key user's desk, configured, governed, and trusted. An ecosystem is still assembling around enterprise AI — particularly for mid-size organizations, where the adoption paths (partner-led, in-house, or vendor-managed) are only now taking shape.

But look at what's loaded behind that calm. The consulting pyramid earns its margin at the base: juniors and offshore capacity, billed at a multiple of cost, doing the standardized work — code adaptation, documentation, test scripts, data migration runs. Now look at what the vendors are shipping. Among SAP's announced agents for 2026 is a mass custom-code conversion agent for S/4HANA migrations, targeting on the order of 40% efficiency gains in exactly that adaptation work. Read that sentence the way an integrator's CFO would. The first agents to mature are aimed, precisely, at the layer of work that funds the pyramid.

The disruption isn't here yet. It's armed.

What the other side of the shift looks like

I can tell you what an AI-augmented practice looks like, because I run one. As an independent consultant, I've broken my own delivery down into projects inside my LLM environment: expertise research, migration analysis, effort estimation, documentation drafting, test generation. Work that used to mean either my evenings or a junior resource now runs through an assistant I direct. What I actually sell has shifted up a level, toward judgment: which option to take, what the risk really is, what the client should refuse.

One consultant reorganizing his own toolbox is an anecdote. But it's a preview of the mechanics: the standardized layer compresses, the judgment layer concentrates, and the person in the middle either moves up or gets priced like the tooling. What I've done alone, client IT departments will do at scale the day the agent tooling lands under license on their side of the table.

The repricing, when it comes

I'll be careful here, because this is the part of the argument where certainty would be dishonest. Nobody can tell you the date, and the consulting firms are not passive — their margins depend on the base of the pyramid, which means they cannot afford to concede it, and they will fight to repackage it before clients unbundle it.

But the danger for them is real, and the mechanism is simple to state. The day AI agents are deployed under license inside client IT departments, not as demos but as governed, configured tools, clients will start comparing line by line. An agent that converts custom code against a junior consultant who converts custom code. An agent that drafts test scripts against an offshore team that drafts test scripts. Some of those comparisons will favor people. Many won't. And every one that doesn't erodes the margin layer the integrator model is built on. The offshore version of this argument is even sharper: a model whose entire pitch is "the same standardized work, cheaper" has no answer to a tool that does the standardized work at near-zero marginal cost.

What appears on the other side of that repricing is a new premium. Not for knowing the transaction codes, since that knowledge is being commoditized, but for knowing how to work the agents: how to direct them, verify them, govern them, and translate between what the business needs and what an agent can be trusted to do. The consultants who can talk to agents will price at a premium. The ones who competed with them won't.

What an IT leader does with this

If you run an IT organization, the practical conclusion is not "wait for the ERP to transform." It's to act on the layer that is actually moving.

Read your integrator contracts through this lens: how much of what you're paying for is standardized work that agent tooling will commoditize over the life of the contract, and what happens to the rate card when it does. Ask your SI, today, how their delivery model incorporates these agents and whether the efficiency gains show up in your price or only in their margin. Rethink the offshore line item with the same question. And start building, internally, the skill that will carry the new premium — people who can direct and govern agents in your landscape — before that skill becomes scarce and expensive in its own right.

Read the market in your invoices

So, where is the ERP market heading? The market itself keeps answering with product roadmaps, and that's the misdirection. The transactional core will change slowly, by design. The economics of everyone around it — integrator pyramids, offshore models, internal teams, independents like me — are the part that's already in motion.

Don't read AI's impact on your ERP in the keynotes. Read it in your invoices, your rate cards, and your service contracts. That's where the disruption will land first. By the time it's visible there, the repricing will already be underway.