Our AI policy
How we use AI ourselves.
We advise organisations on AI policy. So it is only fair to make our own policy public. These are the rules our work, and our own agent fleet, live by.
The principles
Five rules our work lives by.
These are principles, not a system description. What our way of working looks like technically is on the agents page; this is what that way of working has to abide by.
A human approves what goes out
Every text, post or email that leaves our work has been approved by a human. Agents propose; humans decide.
Client data only leaves with consent
Client names, quotes and cases only become public after explicit consent. No consent, no publication, however good the story.
Minimal permissions
Our agents work with as few permissions as possible: they can add and propose, but cannot delete the client archive. Irreversible actions require a human.
Everything is traceable
Every autonomous run is logged. If something fails, we see it the same day, and we see what happened. Without that visibility, nothing runs autonomously at our agency.
European ground where possible
This website runs on European infrastructure, and step by step we are moving more of our own stack to European vendors. Digital sovereignty is not a slogan for us; it is a move in progress.
The boundaries
What we do not do.
- We do not train AI models on client data.
- We publish nothing about clients without consent, including anonymised-but-recognisable.
- We do not let agents take irreversible actions without human approval.
- We do not hide that we use AI: our own agent fleet is right on this site.
Version July 2026. This policy grows with our practice; the current version always lives on this page. Questions about it? Feel free to ask
An AI policy for your organisation?
We make manifestos like this for clients too: guardrails that give room instead of slowing down.