ERP has long been described as the heavy machinery of business, massive systems that encode how money, materials, and compliance are tracked. For decades, they have embodied centuries of accounting practice, tax law, and logistics in rigid, auditable software. But with AI now orchestrating workflows, surfacing insights, and even replacing traditional interfaces, the question shifts: what parts of ERP truly endure, and which will be reshaped or discarded?

I read this article this morning

https://computersweden.se/article/4043494/vilka-delar-av-erp-kommer-att-finnas-kvar-nar-ai-tar-over.html

Its in Swedish but I think you can use AI to translate it ;)

My key take aways from that article with some of my own personal perspective is

1. AI can solve some manual lifting that ERP previously hasn’t been able to solve
   
2. AI can customize functionality to cater for a business competitive edge which hopefully is not part of the standard product
   
3. Non modular and closed ERP systems will probably not be here in a couple of years
   
4. ERP systems will still be the source of source of truth for AI layers built on top of it

I have had the following thoughts which related to point 3 and 4 above.

Rules vs AI

At its foundation, ERP is still very much rule-based.

Double-entry bookkeeping, VAT calculation, payroll tax, depreciation schedules, FIFO/LIFO costing, bill of materials explosions.

These are not problems of prediction but of compliance. They must be deterministic and auditable. In these domains, rules are cheaper, safer, and more reliable than neural nets.

AI thrives in fuzzier spaces:

Anomaly detection, predictive demand planning, supplier risk scoring, or conversational interaction with complex systems.

You don’t want a stochastic model deciding whether VAT is 25% or 6%. But you do want one that flags which customers are most likely to pay late, or which purchase orders deviate from past patterns.

This sets up a layered future which is talked about in the article:

• ERP core: deterministic, rule-based compliance kernel.
  
• AI orchestration: natural language intent translation, anomaly detection, predictive decision support.
  
• AI interface: chat, voice, and AR-driven systems that dissolve the traditional ERP UI.

The ERP Kernel concept

Think of an ERP kernel as the business equivalent of an operating system kernel: a small, rigorous, auditable core that encodes the non-negotiable mechanics of business operations.

Above it, AI agents handle interpretation, prediction, and interaction.

What belongs in such a kernel?

• Ledgers: immutable journals for finance, inventory, assets.
  
• Rule engines: taxation, costing, depreciation, currency conversion.
  
• Canonical data models: organizations, products, documents, currencies.
  
• Contracts: a small set of commands (e.g., PostJournal, ReceiveGoods), deterministic queries, and audit events.

And what stays out?

• UX, workflows, task assignments.
  
• AI-driven decision-making, anomaly detection, forecasting.
  
• Custom organizational policies.

The kernel is deliberately boring: deterministic, explainable, and versioned. Every computation is reproducible byte-for-byte. AI sits on top, turning human intent into kernel commands, and highlighting exceptions.

One source of Truth

Today, every ERP vendor encodes its own slightly different implementations of inventory allocation, depreciation, and VAT. AI exposes the redundancy: why maintain dozens of subtly different rules when one open, audited, “ERP kernel” could serve as the global source of truth? Companies don’t need SAP’s and Microsoft’s separate implementations of FIFO; they need one deterministic module that AI agents can call.

This resembles the shift databases made decades ago: few of us implement storage engines anymore. We trust Postgres or Oracle. ERP may collapse the same way, into a standardized rule kernel plus AI-driven orchestration layers.

Standardization and the question of authority

We don’t need several sources of truth for standard stock picking strategies, for how invoices are defined, or for the basics of bookkeeping. These should be defined once, clearly, and maintained with the same rigor as accounting standards themselves. The big question is: who becomes the canonical keeper of these rules?

Will regulators, standards bodies, or open-source foundations step forward to hold the first “ERP kernel”? Or will one vendor make the leap from proprietary silos to become the authoritative source of truth for critical business mechanics? The transition away from closed ERP systems opens the door for a new kind of standardization, but it is far from clear who will seize that role.

What this might mean for Vendors

If this kernel future emerges, ERP splits into two industries:

1. Kernel keepers: maintaining compliance, auditability, and jurisdictional rule packs.
   
2. AI orchestrators: providing multimodal, context-aware interaction and automation.

The kernel side commoditizes, while the orchestration side becomes the differentiator. The “ERP you interact with” may not be SAP or Dynamics, it may be an AI assistant backed by an open kernel.

Closing thoughts

ERP was historically about enforcing consistency and reducing human discretion.

AI is about reintroducing flexible, context-aware judgment.

The future is not one replacing the other, but a fusion: ERP provides the deterministic skeleton, AI the adaptive nervous system and sensory organs.

ERP will become invisible infrastructure. The human experience of ERP will be mediated by AI: conversational, predictive, multimodal. But the kernel, the boring, deterministic rules, remains indispensable. In fact, they may be the only part of ERP that survives intact.