---
title: "System of Action: what replaces the ERP stack"
url: "https://barbairians.com/blog/system-of-action-replaces-the-erp-stack"
description: "Systems of record store what happened. Systems of engagement collect it. A system of action decides and executes — which is why the old stack boundaries stop making sense."
---

# System of Action: what replaces the ERP stack

July 2, 2026·3 min read·BarbAIrianS

Business software has spent forty years perfecting two categories. Systems of record store what happened — ERP, accounting, WMS. Systems of engagement collect it — CRM, helpdesk, forms. Between them sits the part nobody automated: a human reading both, deciding what to do, and typing the result back into one of them. That gap is what an AI-native system of action closes, and closing it dissolves the boundaries the old categories were built on.

## Why the stack was split in the first place

Not because operations are naturally divided into eleven products. Because integration was expensive.

When connecting two systems meant a project, it made sense to buy a suite per domain and accept that the joins would be done by people. The org chart followed the software: a finance team, a warehouse team, a sales team, each fluent in their own product and communicating by exporting spreadsheets to each other.

The cost of that split was never in the licences. It was in the humans acting as the integration layer — and in the delay between something happening and anyone deciding about it.

## What “action” adds

A system of action is not a chatbot on top of your ERP. It has four properties the old stack does not:

**It reads everything.** Not one domain — orders, stock, cash, contracts, mail, the operational database. The join happens inside the system rather than inside a person’s head.

**It runs unprompted.** It does not wait for someone to open a report. It watches state, and when reality drifts from plan it starts working.

**It decides, with a rationale.** Not “here is a chart”, but “here is what I would do, this is the evidence, this is the confidence”. A recommendation you can accept, reject or amend.

**It executes what it is allowed to.** Within explicit policy: draft the order, prepare the reallocation, open the case. The boundary between suggest and execute is configuration, not architecture — which is why the policy layer is the important part of the design, not the model.

## The uncomfortable implication

If the system reads across domains and acts, then domain-shaped products stop being the natural unit. You do not need a CRM, a WMS and an ERP that each hold a fragment of the same order — you need one place where the order lives and one system that acts on it.

This is why “AI-native” is not a marketing prefix on existing suites. A suite with a copilot bolted on is still eleven data models and a human doing the joins, with a language interface on top. AI-native means the data model was designed for a system that reads and acts across the whole business, and the language interface is a consequence rather than the product.

## What this does not mean

It does not mean firing the finance team and letting a model run the company. In every implementation worth respecting, the pattern is the same: **the machine prepares the decision, a human owns it.** The gain is not fewer people — it is people spending their week deciding instead of assembling.

It also does not mean rip and replace. The realistic path is to put the system of action above what you already run, in read-only mode, until it is demonstrably right more often than the process it is shadowing. Systems of record are good at being systems of record. They were only ever bad at deciding.

## Why it matters for sovereignty

Here is the connection to everything else on this site. A system that reads across your entire business and acts on it is the most sensitive software you will ever run. It sees the contracts, the margins, the salaries, the exceptions.

That is not a system to rent from someone whose terms of service can change. If the system of action is what turns your data into decisions, then owning it — the models, the policies, the infrastructure it runs on — is the difference between a company that has been automated and a company that has been captured.

-   AI-native OS
-   ERP
-   System of Action
-   agents

## Keep reading

-   [The barbarians are already inside the gate](https://barbairians.com/blog/barbarians-at-the-gate-of-enterprise-software)
-   [Sovereign AI is procurement, not patriotism](https://barbairians.com/blog/sovereign-ai-is-procurement-not-patriotism)