Solutions
An AI agent for companies
Short answer
An AI agent in a company reliably handles the work around decisions today — reading a document and pulling data out of it, sorting mail, drafting an answer or a piece of material, finding something across systems. It does not reliably handle decisions with consequences: what to approve, what to pay, what to send out. So we build it to prepare the material and let a person press confirm. It is deployed one task at a time, not as an “agent for everything”.
01What an AI agent in a company actually does
Repeated work with a written brief and a clear input.
- 01Reads incoming documents — invoices, orders, contracts, e-mails — and extracts the data into a system.
- 02Sorts and routes mail — whose it is, what priority, what needs a reply.
- 03Drafts a reply or a piece of material from your own sources, for a person to confirm.
- 04Answers repeated questions — from the site, internally to colleagues — only from what is actually in its material.
- 05Finds something across systems: where an order is, which client, what state it is in.
02Where the line is
The model pulls information out and prepares the work. It does not calculate prices, decide approvals or send anything out without a person seeing it — those are steps with consequences, done by code that returns the same result every time, or by a person.
The second line is certainty. When the agent has no answer it should say so and hand over a contact, not invent one. A useful agent is one whose output you can check, not the one that sounds most confident.
03How to deploy it and what it costs
With a fixed-price audit. A few days spent looking at where the manual work with documents and mail is, ending in a document listing what can be given to an agent and roughly what it costs. The document is yours even if you do not continue.
Then one task, on its own — usually document extraction or mail sorting. The first thing that genuinely runs is usually live within a few weeks. A standalone single-task agent is much cheaper than a custom system and works as a first step.
Frequently asked
- Will an AI agent replace an employee?
- It takes the repeated part — retyping from documents, sorting mail, looking up data. Decisions and disputed cases stay with a person; the agent prepares the material for them.
- Where does our data run and is it fine for GDPR?
- It can be built so the data stays in the EU and the model does not train on it. Exactly what goes to the model and what stays in the system is part of the design — we write about it in the article on GDPR and AI in a company.
- What does it cost and how long does deployment take?
- A single-task agent is typically in the low thousands of euros and a few weeks. More tasks and more integrations get close to a custom system in price and are deployed in pieces.