From a single agent to a network: what happens when AI drives your processes?

A chat assistant in your ERP that answers questions and summarizes documents is useful. Most organizations today are at this level — whether they work with Exact, AFAS, Microsoft Dynamics 365 Business Central, or another ERP — and the initial reactions are positive. People save time, registration becomes more consistent, and data is finally findable.

But this is the preparatory step. The real shift — which you as a director or manager will face within a year, whether you approach it strategically or not — is that you won't work with a single agent, but with a network of agents that together execute entire processes. AI orchestration.

Where one agent stops

Take a quotation process. One agent can read an incoming request, look up the customer in your ERP, retrieve historical orders, and prepare a draft quotation. Good. But then it stops. Someone has to check the quotation, send it to purchasing for a margin scan, send it after approval, follow up after a week, convert it into an order upon agreement, and synchronize that order with production or inventory.

In the current situation, the entire chain after step one is manual work again. What is coming now: for each step there is a specialized agent, and there is an orchestration layer that decides who does what — and when a human intervenes.

What multi-agents look like in practice

A purchasing agent checks margins based on your purchasing conditions in Exact. A sales agent drafts a personalized email based on the customer history in AFAS. A logistics agent asks your WMS if the delivery time is feasible. A finance agent validates in Business Central whether the credit limit allows this.

This is not science fiction. It is a series of small, specific agents that can each perform one task — retrieving, comparing, creating, alerting — and which are steered together by an orchestrator. The orchestrator knows whose turn it is and when, where data flows from one to another, and — crucially — at which moments a human must decide.

Human in the loop, always

This is where many AI stories derail. "Full automation" sounds impressive, but does not work in a real business. Not because the technology can't handle it, but because decisions have consequences: an incorrectly sent quotation costs money, a flawed credit note costs trust, an error in an order costs inventory.

In a well-designed multi-agent process, it is consciously recorded where humans decide. Not everywhere — otherwise you have gained nothing. But at the points where it matters: final transmission to the customer, financial obligations above a threshold, deviations from standard conditions. The agents do the preparatory work, the human signs off.

What this changes for you

As a director of a medium-sized company, here lies the real question: not "which AI tool do we buy", but "which processes in my organization are so repetitive that an agent network can execute them, and where do I retain human judgment?"

For IT and operations managers, this means: stop thinking in terms of one big solution. Start with one agent that proves it works, add a second that takes over the next step in the process, and build a chain that grows step-by-step. Each agent must be individually manageable, individually replaceable, and individually measurable.

Where this goes wrong

On one point: no one supervises the integrated process. You can buy AI agents as separate components. A working network of agents, however, is a design issue — and that is about more than just technology.

Curious about what an agent network would look like for your processes? We help organizations grow step-by-step from a single assistant to a working process network, where the human always remains in control.