AI agent development for recurring work.
An AI agent takes on one specific job, such as posting purchase invoices or answering questions about delivery times. I build it into the software you already use, and above a limit you choose, it waits for your approval.
1,250 euro excl. VAT✦Money back if the report finds no opportunity that pays for itself within a year
AI agents in short
Diginesso builds AI agents for small and mid-sized businesses in the Netherlands, Germany and Italy, and for companies elsewhere that do business in the EU. An AI agent is software that handles one recurring job inside the systems you already have, for example posting purchase invoices from your inbox. Diginesso is the sole proprietorship of Wesley Beerendonk. You start with an AI Process Audit at 1,250 euro excl. VAT, which sets out which job an agent can take over and what that is worth per year.
- Sole proprietorship of Wesley Beerendonk
- Fixed price per phase
- By default: database and application in Frankfurt
- Works in English, Dutch, German and Italian
The basics
What is an
AI agent?
An AI agent is software that handles a recurring task from start to finish in your own systems. It reads what comes in, even when the layout is different every time. Then it checks it against the data you already hold and takes the next step, such as preparing an accounting entry or replying to a customer.
How it differs from what you may already use
- A chatbot
- Answers a question in a chat window. The work that follows, such as posting the invoice, is still yours to do.
- Rule-based automation
- Does exactly what you set up in advance, such as turning a webshop order into an invoice in your accounting software. If anything deviates, it stops.
- An AI assistant you open yourself
- Like ChatGPT or Copilot: you type a question and copy the answer over to where it belongs.
An agent can also read the exceptions, and it enters the result in your own software itself.
Where an agent fits
Which tasks can an AI agent take over?
Work that comes up again and again, where someone currently retypes data from one screen into another. Below are examples by department. I work out whether it pays off for you in the AI Process Audit.
- Finance
Reading purchase invoices from your inbox, matching them to the purchase order and preparing the posting, for example in Xero or QuickBooks.
Invoice processing with AI - Customer service
Answering questions about delivery times or returns using the data from your online shop and your orders, evenings included.
Tell me which questions come in - Purchasing
Pulling the new prices out of supplier price lists, each in its own layout, and preparing them as a proposal. An integration then passes them on.
Connecting your webshop and marketplaces - Sales
Picking up an order from a plain email, where the customer describes the item in their own words, and setting up the order. Scheduling and sending a confirmation then run on fixed rules.
Automating order processing - Production
If your machines already have sensors, an agent can monitor the readings and warn you when a pump runs hotter or vibrates more than usual.
Examples by sector
When there is nothing to read or weigh up, rule-based process automation is enough. For work that no off-the-shelf software covers, there is custom software.
Limits
What an agent does well, and where it stops
An agent is good at reading and sorting, even when things are messy. There are also things you should not expect of it.
What an agent can do
- Read emails, PDFs and scans, even when every supplier uses a different layout or a scan is only half legible.
- Compare data, such as an invoice against its purchase order, and flag any discrepancies.
- Prepare work in your own software, from an entry in your accounts to a reply to a customer.
What it does not do
- It is not error-free. An AI model can say something that sounds convincing and is still wrong, so there are checks around it.
- It cannot reach data outside its job: an agent for your purchase invoices has no access to your personnel files.
- With purchase invoices, the most the agent does is prepare the payment. You approve it in your own bank.
How I build an agent
How does an AI agent stay reliable when it runs every day?
Small error rates add up. Say an agent gets each step right in 95 out of 100 cases. Then a job of ten steps goes right from start to finish in only about 60 out of 100 cases. That is why I build these five things around it.
Your approval above a limit
Above an amount or a level of uncertainty that you choose, the agent does not act on its own. It prepares the item and waits for you. You can make that limit stricter at any time.
Only the access the job needs
Every connection gets the narrowest access that still lets the work get done. Every access to the system, by a person or a program, is logged.
Measured on your own work first
Before the agent goes live, I use examples from your own work to measure how often it gets things right. We agree a target score in advance. If the first version misses it, we stop and you pay only for that phase.
Alongside the manual work at first
For the first few weeks the agent runs alongside your old way of working. The manual process is only dropped once the numbers hold up.
Log, monitoring and a stop button
Every action goes into a log, so afterwards you can see why something happened. The agent is monitored, and one button stops it.
On an ordinary working day
An invented example: a wholesaler with supplier price lists
Price lists that every supplier sends differently
Suppliers send new prices as a PDF or just in the body of an email, and each one does it their own way. Someone picks out the changes and retypes them into the online shop, two marketplaces and the stock list. That takes two hours, and halfway through nobody knows which price is right any more.
The agent reads each price list, looks up the items in your own item list and prepares the new prices as a proposal. Any item it cannot match with confidence is set aside for you to check. After your approval, an integration passes the prices on to the webshop, the marketplaces and the stock list. What took two hours becomes about ten minutes of checking.
A second agent can then pick up the purchase invoices from the inbox, with access to that inbox and the accounts only. Done by hand, each invoice costs between 4 and 15 euro in time.
This wholesaler does not exist. The figures are general benchmarks, not figures from Diginesso clients. The same goes for the payback time of a project like this, which is usually between 3 and 6 months.
What it costs
What does an AI agent cost?
There is no standard price for an agent, because the jobs differ too much. An agent that handles one type of email is a very different job from one that monitors your whole production line. So it starts with an audit, and after that you know the price of each phase before that phase begins.
AI Process Audit
1,250 euroexcl. VAT, usually 1 to 2 weeks
I work out which job an agent can take over and calculate what each opportunity is worth. If the build assignment is agreed within 3 months of the report date, the full amount comes off the first invoice of the build.
If the report names no opportunity whose calculated benefit over twelve months exceeds 1,250 euro, you get your money back. An email is enough, up until you approve the report. The report counts as approved if I hear nothing from you 14 days after sending it.
Build
Per phasefixed price, agreed in advance
The report gives an estimate of the build cost for each opportunity. What I build specifically for you becomes yours once the invoices are paid. My own standard components stay mine, but you get a perpetual, transferable licence to use them.
Maintenance
Per monthfixed amount, one month’s notice
I monitor the agent and fix anything in my own work that breaks, at no extra charge. Work needed because you changed something yourself in a connected system is agreed separately in advance. You pay for hosting, the database and AI model usage yourself, on your own accounts. Any extension is priced separately up front.
How I work sets out the four phases of a project. Everything we agree is in the terms of business.
The builder
Who builds your AI agent?
I am Wesley Beerendonk. I am Dutch and live in Riparbella, a village near Pisa. I build the agent myself, and I keep it running afterwards. For specialist work, such as data analysis or design, I work with specialists. You will know in advance who is involved.
More about meQuestions
Frequently asked questions about AI agents
How much does it cost to have an AI agent built?
Building an AI agent comes at a fixed price per phase, agreed in advance. You start with an AI Process Audit at 1,250 euro excl. VAT. For each opportunity it includes an estimate of what the agent costs to build. I only do extra work outside a phase once you have agreed to the price in advance.
What does an AI agent cost per month?
An AI agent costs a fixed monthly amount for maintenance, with one month’s notice. On top of that you pay for hosting, the database and AI model usage yourself, on your own accounts. We agree the monthly amount in advance, and I do not raise it without your consent.
How is an AI agent different from Zapier or Make?
Zapier and Make are integration tools that run on fixed rules you set up in advance. That works well as long as everything looks the same each time. An AI agent can also read free text and deal with exceptions, such as an invoice in a layout it has not seen before. That makes it less likely to get stuck than a fixed rule.
What if an AI agent makes a mistake?
An AI agent can make mistakes, so a person stays between the agent and any decision where a mistake could do damage. I do not guarantee that every output of an AI model is correct. What I do is build in checks and measure in advance, on examples from your own work, how often the agent gets it right. The handover states in writing what quality was measured and in which cases the agent gets it wrong.
Where is my data when an AI agent works with it?
In a standard project your database and the application sit in Frankfurt, and the emails the system sends go out from Ireland. You choose the AI model; with a European model your data stays within Europe. Supabase, Vercel and Resend are American companies with servers in Europe. If you would rather avoid that, there is a setup with European providers only. If personal data is involved, you sign a data processing agreement under the GDPR before any records are touched. Every supplier is listed by name and region on the Secure and European page.
How long does it take to build an AI agent?
Allow 5 to 12 weeks from the start to an agent that runs every day. The AI Process Audit takes 1 to 2 weeks, the first version on your data 2 to 4 weeks, and the live and monitored phase 2 to 6 weeks. After each phase you decide whether we carry on.
Who owns the agent once it is finished?
An agent built by Diginesso is yours once the invoices are paid. That covers everything I build specifically for you, such as the code, the integrations and the instructions for the AI model. My own standard components stay mine, but you get a perpetual, transferable licence to use them. The accounts for hosting, the database and the AI model are in your name, so if I am no longer around, the agent keeps running.
What does the EU AI Act say about AI agents?
For ordinary business tasks, such as processing invoices, the EU AI Act sets few requirements. Since 2 February 2025, though, you have to promote AI literacy among the people who work with it; a short briefing is usually enough. If a customer talks to your agent, since 2 August 2026 that customer must be able to see they are dealing with AI. That is in Article 50. High-risk systems, such as those used in recruitment or credit scoring, face stricter requirements from 2 December 2027. That date moved because of the Digital Omnibus, Regulation (EU) 2026/1744. I am not a lawyer, so have borderline cases checked by one.
Which job would you hand over first?
Describe it in a few sentences. You get a personal reply from me within one working day. Then we will both know whether an agent is worth it for that job.