AI assistants promise a lot, but the moment you ask for anything beyond a chat, you see where they fall short. "Create a sales order, check stock, schedule delivery and send the invoice" — that is not a single question, but a chain of actions across your entire Exact Online administration. One AI model trying to do it all on its own quickly trips over the details.
The answer lies in two concepts you increasingly see in modern AI platforms: Skills and the Agent of Agents architecture. Together they make it possible to truly automate complex ERP processes — from quote to payment — with AI. In this article: how it works and what it concretely delivers for Exact Online users in SME businesses.
What are Skills in AI?
A Skill is a bounded, reusable capability that an AI agent can invoke to perform one specific task. Compare it to kitchen tools: a knife for cutting, a whisk for whisking, an oven for baking. No single tool does everything — but together you get a three-course meal.
Concrete examples for your administration:
- An inventory Skill that fetches current stock levels and processes warehouse transfers.
- An order management Skill that creates or updates sales orders, deliveries and returns.
- An invoice Skill that manages debtors, fetches outstanding items and registers payments.
- A production Skill that drives shop orders, materials and bills of material.
Each Skill is small, predictable and easy to test. The AI agent itself does not need to "know" how your ERP works — that knowledge lives inside the Skill.
What is an Agent of Agents?
An Agent of Agents is an architecture where a coordinating main agent receives questions and dispatches execution to specialised sub-agents. Each sub-agent has its own set of Skills and a bounded domain.
Think of a conductor with an orchestra. The conductor reads the score (your question) and signals which section plays when. The strings, brass and percussion do not need to know what the others are doing — they only need to trust the conductor and play their part.
In a multi-agent system for your administration, this looks like:
- The main agent interprets the user request in natural language.
- An inventory agent handles everything related to warehouses and items.
- A logistics agent manages deliveries, receipts and transport.
- A financial agent handles invoices, payments and journal entries.
Each agent does what it is good at, and the main agent ensures outcomes are consistent before reporting back. That scales far better than one all-rounder AI trying to do everything at once.
Skills + Agent of Agents in Exact Online
Make it concrete. Suppose you ask your AI assistant:
"Send 50 units of A1001 to Baker De Groot and create the invoice."
A traditional AI tries this as one command — and gets confused about which steps go in what order. In an Agent of Agents approach, this is what happens:
- The main agent decomposes the question into three sub-tasks: stock check, delivery, invoice.
- The inventory agent uses its Skill to verify 50 units of A1001 are available.
- The logistics agent creates a goods delivery and links it to the correct sales order.
- The financial agent generates the invoice with the right VAT code, customer details and payment terms.
- The main agent validates that all steps were completed without errors and returns a confirmation.
Under the hood all those agents talk to your Exact Online administration via the Model Context Protocol (MCP). Ledger Botje provides that bridge: it exposes 220+ Skills your AI assistant can call, from searching items to finishing a production order.
Three real-world examples
1. Order-to-cash at a wholesaler
A customer calls for an urgent order of 200 items. Your AI assistant:
- Checks stock in your warehouse (inventory agent).
- Creates a sales order at the correct price (order agent).
- Reserves the goods and schedules delivery for tomorrow (logistics agent).
- Sends the invoice directly to the customer (financial agent).
Without an Agent of Agents you would have had to click through four different Exact Online screens manually. Now it is one natural-language sentence.
2. Procure-to-stock at a manufacturer
Stock of a component drops below the reorder point. The AI assistant:
- Detects that A1001 has dropped below reorder point (inventory agent).
- Looks up the preferred supplier and fetches the current purchase price (procurement agent).
- Creates a purchase order and submits it for approval (order agent).
- Schedules the expected receipt so production knows when components arrive.
What normally takes half a workday of administration becomes a two-minute request.
3. Month-end close at an SME administration
At the end of the month:
- A reporting agent fetches all outstanding items.
- A payments agent checks which invoices have been paid and which need a reminder.
- A general-ledger agent verifies that all journal entries are coded correctly.
- The main agent delivers a summary with action items.
Your accountant or bookkeeper no longer starts the month-end with sorting — they start with deciding.
Why this matters for SME businesses
A multi-agent system is not a tech parade. For SME businesses using Exact Online daily, the Agent of Agents approach delivers four concrete benefits:
- Fewer errors: each Skill is specialised and tested. An invoice Skill knows VAT codes, an inventory Skill knows warehouse locations — less goes wrong than with manual re-typing between screens.
- Scalability: when you grow from 50 to 500 orders a day, the workflow stays the same. The agents simply keep working without you having to onboard extra staff.
- No manual re-typing: the AI fetches data directly from your Exact Online administration and writes data directly back. Intermediate steps via Excel, email or printed paper disappear.
- 24/7 availability: customer questions after hours? A routine task on a Sunday? The Agent of Agents stands ready without overtime or weekend pay.
For an average bookkeeper this means: more time for analysis and advice, less time clicking through seventeen different Exact Online screens.
Getting started with Ledger Botje
Ledger Botje is the MCP connection between Exact Online and AI assistants like ChatGPT, Claude, Microsoft Copilot, Gemini, Mistral, Perplexity and Grok. Under the hood Ledger Botje acts as the Skill layer: 220+ predefined tools your AI assistant can invoke to do something in Exact Online.
You do not have to program anything. The connection works via the Model Context Protocol — an open standard supported by OpenAI, Anthropic and others. See the MCP installation guide for how to connect Ledger Botje to your favourite AI assistant in five minutes. In the prompt library you will find examples for the most common scenarios, from quote to month-end close.
Conclusion
Skills give AI assistants a bounded capability. An Agent of Agents architecture lets all those capabilities work together on complex tasks across your administration. For Exact Online users this means: you describe a desired outcome in plain English, and behind the scenes a team of specialised agents performs the steps.
Want to know what automating Exact Online with AI delivers for your business? Browse the Exact Online AI integration, read more about AI accounting, or try Ledger Botje free for 14 days. No credit card needed.