
AI agents
The same craft as the rest of what we build, except that this piece of software answers back.
- Part of
- Development
- Related to
- AI Transformation
- Live example
- Top right of this site
- First version
- Within weeks

An agent is software, not a subscription.
That sounds obvious. Yet a lot of what is on offer in this corner is a monthly fee for a chat window that reads your own website. Handy, except it knows nothing about your quotes, your stock or your agreements.
What we build is tied to your systems. That makes it a piece of software you maintain, extend and own, like the rest of what we make for you. The proof is in the top right of this page: we built that assistant for ourselves.
Where this was part of the work
01 / 21
The model is the easiest decision in the whole project.
The work is in the rest: which sources it may consult, how it knows which version of a document is valid, what it does when two systems contradict each other, and when it hands over to a person.
That is software development, with the same demands as any other system: testable, monitorable, changeable. Skip that and you build a demo.
Brands we build this for
It has to be able to say it doesn't know.
An agent that always gives an answer will sooner or later give a wrong one with the same confidence. So we set out in advance what it covers, which sources it may use and when it refers on.
The same goes for what it sees: which data the system may consult and what does and does not go to a model, we agree before anything is built.
Common questions
Not answered here? Ask the AI in the top right, or call us.
In substance nothing; it is the same technology and often the same team. Under Development the emphasis is on the engineering: integrations, data model, maintenance. Under AI Transformation it is on what it changes in your organisation and how you get it adopted.
If that system has an API or an export option, yes. The integration is often the largest part of the work, and it decides whether the agent becomes useful or merely pleasant.
By having it answer from your sources rather than from the model's memory, and by setting out what it covers. On top of that we watch what gets asked, so gaps and misunderstandings show up quickly.
A first working version on a bounded subject is in the order of a normal build project of a few weeks. On top of that come the usage costs of the model and ongoing maintenance. You get a range up front, including the points we aren't sure about yet.
It can, and sometimes it has to. Where the data lives and which model may see it, we decide together beforehand. If you work with sensitive information, we choose the setup accordingly rather than afterwards.
What it produced
+0%
more orders
at Smart Charged
+0%
more enquiries
at Streetgasm
0
specialists on the team
0+
collaborations
The list of questions is the most valuable thing that comes out.
Every month you see exactly what visitors or colleagues want to know and where the agent runs aground. That is free research into what is missing from your site or your process. Usually it leads to changes that have nothing to do with AI.
That is why an agent is never finished. Your knowledge changes, your offering changes, and the questions change with them.
Which question do you get most often?
That's where an agent with a subject starts.
Tell us which questions come in most often, and which systems it would have to talk to. In half an hour you'll hear what it takes.




















