AI Leadership and the Future of ERP
AI will not replace ERP over the next three to seven years. What it will do is something subtler and, in the long run, more consequential. It will loosen the ERP system's grip on how people actually work, on where decisions get made, and on who captures the economic value that surrounds enterprise processes.
The real question is not whether ERP survives AI. It is which layer of the enterprise stack will own the next generation of work: the system of record that governs transactions, or the intelligence layer that guides people through them.
Consider first what will not change. The general ledger, inventory balances, production orders, costing rules, approval hierarchies, tax logic, and audit history all depend on deterministic structure. A business must still be able to say, with confidence, that a transaction happened, that it was authorized, and that it can be traced. McKinsey and Deloitte have both arrived at essentially this conclusion: the interaction layer may become agent-driven, and the ERP itself may increasingly fade from view, but the system-of-record foundation remains indispensable for rules, consistency, compliance, and scale.
What is changing is who controls the enterprise experience.
For most of ERP's history, the vendor owned the whole stack: the database, the application logic, the workflow, the analytics, and the screen in front of the user. AI pries those layers apart. Imagine needing the full story on one customer: why their last order shipped late, what they still owe, whether the quality hold on their next shipment is real, and what you should promise them on this afternoon's call. Today that means opening the ERP, then CRM, then the shop-floor system, then a chain of emails. Soon, you may simply ask your own personal enterprise agent. It reaches into ERP, but also into CRM, MES, email, contracts, warehouse systems, and external data. It calls on several specialist agents, assembles a recommended response, and, once you approve it, writes the resulting actions back to ERP. You never open a traditional ERP screen at all.
That, to me, is the most credible form of disruption. ERP remains the system of record, but another platform becomes the system of intelligence, interaction, and orchestration, and with it goes much of the strategic leverage.
The current leadership picture reflects exactly this tension. Among the large incumbents, SAP and Oracle are best positioned, and for good reason: they hold broad process coverage, deep transactional context, hard-won industry knowledge, substantial data platforms, and ecosystems that took decades to build. Oracle today offers some of the clearest evidence of agentic applications that customers can use across finance and supply chain. SAP has the broader architectural vision, and its Autonomous Enterprise portfolio announced in May 2026 is the most ambitious statement of where ERP is heading. As with any roadmap of that scale, executives should track what is generally available today against what is still to come, and plan accordingly.
Microsoft may prove the most strategically disruptive of all, precisely because it does not need every ERP interaction to begin inside Dynamics. Its advantage lies in carrying ERP data and actions into the places where people already spend their day: Excel, Outlook, Teams, Copilot, the Power Platform, and systems beyond its own walls. That makes plausible a future in which Microsoft owns the working interface even when SAP, Oracle, or another vendor owns the transaction. Microsoft's September 23, 2026 roadmap makes that direction explicit through the Finance Agent, ERP access through MCP, cross-system connectors, and reusable business skills.
Some of the more focused ERP vendors have reached the same conclusion from the other direction, by opening their platforms to outside AI rather than building everything themselves. Acumatica's AI Studio lets a customer connect the large language model of its choice, whether from OpenAI, Anthropic, AWS, or Azure, directly to ERP screens with no custom development. IFS has gone further, partnering with Anthropic through its Nexus Black unit to build industry-specific agents such as Resolve for frontline industrial work. Tight relationships with the AI players, rather than homegrown models, are how these vendors intend to stay relevant.
ERP is not going away. What is going away is your relationship with it.
Beneath that top tier, the story becomes one of focus rather than breadth. IFS is unusually strong in asset-intensive operations, maintenance, field service, and bounded industrial workflows. Infor and Epicor pair genuine manufacturing and distribution context with emerging orchestration and agent-building capabilities. None of these firms needs to match SAP or Oracle feature for feature; they can lead where industry understanding matters more than horizontal reach. Some of the most convincing proof points are coming from this tier: IFS can point to named customers already running its agents in live maintenance and service operations, and Epicor can report how many agent requests its customers process in production each month. Real customers and real usage volumes tell an executive more than generic agent announcements alone.
ERP incumbents are far from defenseless. They hold the data structures, permissions, process rules, historical transactions, regulatory support, and customer trust that any external agent must borrow to be useful. External platforms confront an uncomfortable truth here: ERP information is mission critical. An agent that proposes the wrong inventory movement, journal entry, production change, or supplier payment creates operational and control risk. Recommending is cheap; executing is not. The moment an agent stops suggesting and starts posting transactions, it needs the same guardrails a human user does: role-based security, embedded process rules, segregation of duties, and a complete audit trail. Those guardrails take years to build and prove, and the ERP incumbents already have them.
That is why I believe the destination is hybrid or best-in-class, rather than purely native or purely platform, at least for the next three to seven years. After that, the balance may shift again.
For a manufacturing or distribution executive, the practical questions are therefore not "Which vendor has the most AI?" or "Whose demo was most impressive?" The questions that matter are harder and more operational: "Can this system still close the books, ship the order, and pass the audit on the day an agent gets it wrong?" "Can an agent we did not buy from this vendor read our data and act on it, under our permissions and our approvals?" And "If we change our mind about the intelligence layer in three years, what will it cost us to leave?"
Behind those questions sits a set of structural commitments executives should demand: strong APIs, event access, portable data, explainable decisions, thresholds for human control, complete audit trails, interoperability with external agents, and commercial transparency. And they should insist on testing agents against the messy realities of their own business, including shortages, supplier delays, quality holds, cost variances, forecast errors, warehouse constraints, and customer commitments, rather than against polished generic scripts.
So here is my conclusion, and it is deliberately calm. ERP is not going away. What is going away is your relationship with it. The screens will fade, the intelligence will spread across agents we did not buy from the ERP vendor, and implementations should become faster and cheaper. Modules will be unbundled and new competitors will crowd in around data, agents, and orchestration. But underneath all of that motion, the ledger, the transactions, and the audit trail will still be there, quietly doing the work that makes everything else trustworthy. That core will outlast this article, and probably the agents too.
In closing, my advice: choose the core carefully. Keep the architecture open. Govern the data. Design the process. Preserve accountability.