It depends on the work the agent takes over, the systems it works in and where a person needs to stay in the loop. Answer a few questions and you'll get an estimate you can check.
Built for clients including Tata Steel, M&G Investments, Digistore24, Toppers in Concert, Bar-le-Duc and Loogman
Cost check
Work out what your AI agent costs
A few questions, then a request for an estimate you can check. No obligation.
Step 1 of 8Takes about two minutes
What drives the price
What drives the cost of an AI agent
Two agents that both handle email can be very different in size. These are the six things that make the difference.
The work and its exceptions
A request that always follows the same path is quick to set up. The work is in the exceptions: free-text email, documents that differ by sender, requests that need understanding first. The more variations, the more testing it takes.
The number of systems
An agent is only as useful as what it is allowed to do. Reading one mailbox is different from changing orders in the ERP and updating the CRM. For each system, what counts is the quality of the integration and the permissions the agent gets.
Human approval
For each task we agree what the agent may do on its own, what goes past a member of staff first and when it hands the work over. A well-designed approval step takes build time and takes the risk out.
Logging and oversight
Every action is recorded: what the agent saw, what it decided and what it did. That logging, and a place where someone can look over its shoulder, is part of the build, not a later phase.
How sensitive the data is
Personal data or confidential documents call for different choices than public information: which model, where it runs and whether hosting within the EU is needed. That affects what building and running it costs.
Running costs after go-live
An agent also costs money while it runs: language model usage, hosting, monitoring and adjustments when the work changes. That belongs in the budget, not as a surprise afterwards.
From question to quote
How we arrive at a price
1
Your answers
The cost check tells us what you want built, who it is for and which parts it needs.
2
A short call
We call you back, ask what it needs to achieve and what is already in place, and tell you if an off-the-shelf package would suit you better.
3
A quote you can check
Scope, timeline and price on paper before anything is built. If the budget doesn't fit, we show you what to build first.
The questions we hear most often before someone has an AI agent built, with the answers we also give in a first conversation.
1
How much does an AI agent cost?
The honest answer: it depends on the work. An agent that labels incoming email and routes it to the right colleague is a different project from one that reads the request, looks up the order in the ERP, changes it and prepares a reply for approval. The number of systems, the number of exceptions and the points where a person decides account for most of the price.
That is why we don't quote a figure on this page. A price from a pricing model only means something for an agent once you know what the work is, which systems are involved and what data it handles. With your answers from the cost check we give you an estimate you can check, and in the quote we fix scope, timeline and price before anything is built.
2
AI agent or regular automation
Not every process needs an agent. If the rules are fixed and the input is clean, a regular integration or workflow is quicker to build, cheaper to run and easier to manage. We'll tell you that, even if you come to us asking for an agent.
An agent is only worth it when language or judgement is involved: free-text email, delivery notes that differ per supplier, a request that needs understanding first. Often a combination is the most cost-effective: automation for the fixed steps, an agent only for the part that needs reading and deciding.
3
Why quotes for an AI agent vary so much
A demo of an agent is quick to make. An agent that works in your own systems every day, with permissions per system, approval where it matters and logging of every action, is a different product. The difference between quotes is usually in what falls outside the demo.
Ask whether integrations, logging and approval steps are included in the price
Ask how the agent is tested on examples from your own work
Ask what language model usage costs, and who pays that bill
Ask whether you are tied to one model or one provider
4
Start small without throwing work away
The best start is one task for one team. At first the agent runs alongside the team and everything goes past a member of staff; what proves to work well may then run on its own. That way you see what it does before you build further.
The risk of starting small is a proof of concept that ends up in a drawer after the demo. So we lay the foundations that last from day one: access to the systems, integrations, logging and an owner for each step. The second task builds on that instead of starting again.
5
What does an AI agent cost after go-live?
An agent isn't finished at handover. The work changes, systems get updates and exceptions turn up that weren't in the tests. That's why we monitor after go-live, adjust and extend where it pays off.
On top of that you pay for usage. The language model costs money per task it carries out, or in hosting if the model runs on its own server within the EU. How much depends on how often the work comes up, how much text the agent reads per task and which model the task needs. We work it out in advance, so you know where you stand.
6
How to keep costs under control
The most expensive agent is one that does work nobody needed, or that stops working as soon as the builder leaves. So start with the work that takes the most time now, work out what it costs, and only then build.
Start with one task that comes up often, not the most impressive use case
Leave fixed steps to regular automation
For each task, choose the smallest model that does it well
Build model-independently, so you can switch later
Amsterdam
Who you'll talk to
You talk to the founders
No salespeople. We read your answers and call you back. Then you build with a permanent team from our office in Amsterdam.
Frequently asked questions about AI agent development
How much does an AI agent cost?
It depends on the work the agent takes over, the number of systems it works in, the exceptions in the process and where a person needs to decide. The cost check on this page lets you share that in a few minutes, and you get an estimate you can check. In the quote we fix scope, timeline and price before anything is built.
Why isn't there a price on this page?
Because a price from a pricing model is almost always wrong for an AI agent: it only means something once you know what the work is, which systems are involved and what data it handles. We would rather give you an estimate based on your answers, from people who have read them, than a figure that sends you in the wrong direction.
Is an AI agent more expensive than regular automation?
Usually, both to build and to run: an agent has to be tested on exceptions and uses a language model for every task. If the rules are fixed and the input is clean, regular automation is the better choice, and we'll say so. An agent pays off where language or judgement is needed.
What does the language model behind an AI agent cost?
That depends on how often the agent works, how much text it reads and writes per task and which model the task needs. A model via an API is paid per use; an open model running within the EU mainly costs hosting. We are model-independent, choose per task and work it out in advance.
What does it cost to maintain an AI agent?
After go-live we monitor the agent, adjust it when the work or a system changes and extend it where that pays off. Hosting and language model usage come on top. What that comes to each month depends on the scope, and we agree it in advance.
How long does it take to build an AI agent?
That depends on the number of systems and exceptions, not on a standard timeline. The agent first runs alongside one team, with everything going past a member of staff, before it works on its own. The timeline is in the quote.
Who owns the AI agent once it's finished?
You do. The agent, the instructions, the integrations and the code live in your environment, with documentation, and your data stays yours. Because we build model-independently, you aren't tied to one language model provider either.
What if I don't know yet what work the agent should do?
Then start with the AI Opportunity Audit. In it we find out which work takes the most time now, what it costs and whether an agent is the right answer. The audit has a fixed price, known in advance, and may also conclude that nothing needs to be built.
Cookies & privacy. We use functional cookies (always) and analytics and marketing cookies (only with your consent) to improve the site. Read our privacy policy.
