Stop waiting for AI to kill ERP
Agentic AI will not replace ERP. It will expose weak foundations. Learn how clean, governed ERP becomes the execution backbone for intelligent work.
Are you ready to join the AI leaders?
ERP is not dead. But the old ERP strategy is.
For years, many companies treated ERP as a necessary burden: expensive to change, painful to upgrade, but too critical to touch. Now agentic AI has entered the conversation, and some executives are asking whether they should simply wait until AI agents replace parts of SAP, Oracle or Microsoft Dynamics or any other ERP platform.
That is the wrong question.
The better question is: is your ERP landscape clean, connected and trustworthy enough for AI agents to use?
Because agentic AI does not magically fix fragmented master data, undocumented custom code, broken process ownership or weak controls. It amplifies whatever foundation it runs on.
This matters now because two decision windows are converging. Many companies still need to make hard ERP choices: modernise, move to cloud, carve out complexity, extend legacy platforms, or finally address years of customisation. At the same time, agentic AI is moving from demo to workflow, but not yet to broad enterprise maturity.
SAP is one obvious example. SAP Business Suite 7 mainstream maintenance ends at the end of 2027, with optional extended maintenance until the end of 2030, while SAP has committed to S/4HANA maintenance availability until 2040. In parallel, PwC Switzerland’s Cloud Business Survey shows that only 15% of Swiss organisations are implementing or scaling agentic AI. The adoption curve is real, but full enterprise maturity is not here yet.
So the boardroom fantasy that companies can pause ERP investment and wait for agents is dangerous. The companies that wait may discover that their ERP is too fragmented for AI to be used safely, and their AI is too immature to replace ERP.
What agentic AI will actually do to ERP
Agentic AI will change ERP profoundly. But not by replacing the system of record first.
It will first replace the friction around ERP: searching, reporting, chasing approvals, reconciling exceptions, validating invoices, preparing close tasks, answering policy questions, onboarding suppliers and guiding users through complex workflows.
The useful distinction is this: ERP remains the system of record; agents increasingly become the system of action.
ERP should continue to own what must be trusted: ledger, inventory, tax, statutory reporting, approvals, segregation of duties, master data, auditability and enterprise-wide process integrity. Agents should increasingly own what should become faster, more adaptive and more intuitive: finding context, preparing decisions, explaining exceptions, triggering workflows and coordinating work across systems.
This is where the big ERP vendors are already moving.
SAP is positioning around the “Autonomous Enterprise,” combining SAP Business AI, SAP Business Data Cloud, SAP Knowledge Graph, Joule Studio and agent capabilities. Its bet is that agents need deep process context, business semantics and governance to work safely in mission-critical workflows.
Oracle is embedding agents directly into Fusion Cloud Applications. Its Fusion Agentic Applications for finance and supply chain use specialised agent teams that operate inside existing workflows, permissions, approval hierarchies and transactional context.
Microsoft is taking the ecosystem route. Dynamics 365 ERP agents are tied to Copilot, Microsoft 365, Dataverse, Power Platform and Model Context Protocol, allowing agents to connect with finance and operations business logic while users remain close to Outlook, Teams, Excel and Copilot-based experiences.
Different strategies, same conclusion: ERP vendors are not preparing for a world where agents eliminate ERP. They are preparing for a world where ERP becomes the trusted execution layer for agents.
The caveat: CIOs should distinguish vendor roadmap, generally available capability and proven capability in their own control environment. The direction is clear, but many agentic ERP capabilities are still early, uneven and highly dependent on the customer’s architecture, data quality and governance maturity.
The next ERP battle: who owns the agentic control plane?
The real competition is shifting.
In the past, ERP competition was mostly about functional coverage: finance, supply chain, procurement, manufacturing, HR. In the agentic era, the battle will increasingly be about who controls the business context layer and the agentic control plane.
The next lock-in may not be the ERP database. It may be the place where business context, permissions, workflow memory, process semantics, tooling, monitoring and AI orchestration come together.
SAP wants to win with process depth, semantic business data and clean-core discipline. Oracle wants to win with embedded, outcome-driven agentic applications inside Fusion. Microsoft wants to win by making ERP intelligence available in the daily productivity flow and by connecting agents through its broader cloud, data and productivity ecosystem.
This matters for CIOs because vendor choice will increasingly shape agent architecture. A company heavily invested in SAP may benefit from SAP’s process graph, business data layer and clean-core strategy. A company standardised on Oracle Fusion may prefer embedded agentic applications with less integration burden. A Microsoft-centric organisation may move faster by connecting ERP, productivity, low-code and Copilot experiences.
But no vendor removes the CIO’s accountability. Agents will still need identity, controls, audit trails, human approval patterns, data quality, process ownership, security design and cost governance.
Gartner has warned that more than 40% of agentic AI projects may be canceled by the end of 2027 because of unclear value, rising cost or inadequate risk controls. That is not a reason to avoid agentic AI. It is a reason to stop treating it like a playground.
A practical control model for agentic ERP
The question is not only what an agent can do. The board-level question is what the agent is allowed to do, who approved it, how it is monitored, and how the company can prove what happened afterwards.
| Agent level | Typical ERP use | Minimum control logic |
| Observe | Read-only reporting, policy lookup, transaction explanation | Access control, logging, source traceability |
| Recommend | Propose journal entries, supplier actions, stock movements or close tasks | Human decision ownership, rationale, explainability, approval routing |
| Act with approval | Prepare and submit workflows, draft updates, trigger controlled actions | Explicit approval, segregation of duties, audit trail, rollback pattern |
| Act autonomously | Execute bounded low-risk actions within thresholds | Strict guardrails, monitoring, exception handling, periodic control testing |
This is where many ERP strategies are underprepared. Companies have spent years debating cloud migration and customization, but less time defining agent permissions, transactional boundaries, model monitoring, exception handling and accountability for AI-assisted decisions.
My point of view: modernise ERP, but stop worshipping it
CIOs should not choose between ERP and agentic AI. They should divide the battlefield.
Keep ERP for what must be trusted: ledger, inventory, tax, statutory reporting, approvals, segregation of duties, master data, auditability and enterprise-wide process integrity.
Use agents for what should become faster, more adaptive and more intuitive: exception handling, reconciliation, close support, procurement workflows, service resolution, planning scenarios, reporting, user guidance and process orchestration.
Start with use cases where agents reduce friction without weakening controls.
| Risk level | Good starting points | Examples |
| Lower risk | Read, explain, search, summarise | Ask ERP questions, explain invoices, summarise purchase orders, find policy guidance |
| Medium risk | Prepare actions for review | Invoice exceptions, supplier onboarding, close task preparation, reconciliation support |
| Higher risk | Execute controlled actions | Journal posting, payment term changes, releasing blocked invoices, master data updates |
The biggest mistake is to pour money into ERP customisation that recreates yesterday’s workflows. That work will age badly. The better investment is a clean core, strong APIs, reliable data products, modern integration, process mining, governance and selective agentic use cases with measurable business outcomes.
Clean core does not mean no differentiation. It means architectural discipline. Companies still need differentiation, but they should stop embedding every local habit into the ERP core. The more logic sits in undocumented custom code, the harder it becomes for agents to understand, govern, test and safely execute work.
There is also a cost warning. Agentic AI will not be free automation. The cost shifts from user licenses and customisation into orchestration, data products, monitoring, model consumption, testing, exception handling and control assurance. CIOs need AI cost governance before agent sprawl becomes the next shadow IT problem.
The provocative truth is this: agentic AI will not kill ERP. It will kill bad ERP
Over-customised cores, weak data models, broken master data, undocumented process workarounds and disconnected process landscapes will become liabilities. Clean, governed, cloud-ready ERP platforms will become launchpads.
So do not wait for AI to replace your ERP. Make your ERP ready for AI.
The companies that win will not be the ones that wait for AI to make ERP irrelevant. They will be the ones that make ERP understandable, governed, connected and safe enough for AI to act on.
The next ERP strategy is not about worshipping the core. It is about making the core trustworthy enough to become the execution backbone for intelligent work.