The short answer
Having an AI agent built at Tamboer AI costs from 4,500 euro excluding VAT per process and takes three weeks, including testing, training and handover. You get an agent that prepares work inside recorded boundaries, with an employee who approves. If the work is identical every time, you need automation rather than an agent, and automation is cheaper.
What the price covers
| Part | What you get |
|---|---|
| Boundaries on paper | What the agent may read, what it may do, who approves, when it stops and what it records |
| The agent itself | Built on your sources and your examples, running inside the systems you already use |
| Test evidence | Run on real cases from your own archive, with the outcomes side by side |
| Training | Your team learns to judge drafts, handle exceptions and stop or resume the agent |
| Handover | You own it. Support is optional at 125 euro per hour with a ceiling agreed up front |
| Price and time | From 4,500 euro per process, three weeks, or twelve monthly instalments of 375 euro |
Why an agent costs more than it looks
Building is rarely the problem. An agent costs money in two places you do not see up front. First per case: it charges per step the model takes, and how many steps that is varies by case. Second in control: an agent that decides its own order has to record every step, otherwise you cannot check afterwards what happened. Skipping that saves money on the only thing you will later need.
When you need no agent
If someone who knows the work can write down in advance which step follows when, without exceptions, then it is plain automation. That is cheaper, predictable in cost and fully auditable. The difference is explained on AI agent or automation. In practice I often build both inside one workflow: automation to fetch and store, the model for the step that asks for judgement.
And sometimes the answer is neither. If the work occurs a few times per month, you never earn the build back. A good prompt in ChatGPT or Copilot is then enough, and I will say so.
What an agent may never do here
Send, book or commit. The agent prepares and puts work ready, an employee reviews and decides. That is arithmetic rather than caution: an agent fails with a plausible wrong answer rather than an error message. Those costs land on you, never on the model. The five controls that record this boundary sit on the page about control and security.
What you have to supply
- Twenty to fifty real cases from your own archive, anonymised if you prefer.
- The source the agent takes its facts from, and who keeps that source current.
- The name of whoever approves, and what that person may reject.
- The cases you already know go wrong. Those are worth the most.
How to start
With a free process diagnosis: you hear whether an agent fits, whether automation is better, or whether you should leave it alone. Prefer to look first? The free process check returns fit, doubt or no-go in seven questions. The full price build-up sits on the pricing page.
Frequently asked questions
What does it cost to have an AI agent built?
From 4,500 euro excluding VAT per process, including testing on real cases, training and handover, or twelve monthly instalments of 375 euro. On top of that come the running costs of the model itself, which vary per case.
How long does building an AI agent take?
Three weeks from the moment scope and examples are available. Supplying real cases and access decides whether that schedule holds.
May the agent send emails on its own?
Not in the way of working I build. The agent prepares and puts work ready, an employee approves and sends. If you want an agent that acts on its own, I am the wrong supplier.
Do I need an agent or is automation enough?
If you can write the steps down in advance without exceptions, automation is enough and cheaper. If it differs per case and calls for judgement, that is where the agent sits.
