How we work
Built carefully, then looked after.
We are a small studio, so we build fewer things and do them properly. Here is what that means in practice, and the unusual part: who does most of the work.
- 1
Staging first, every time
Every app has a staging copy with made-up data. A change is planned, built and tested there before it goes anywhere near a real person.
- 2
Your AI asks before anything that matters
Our apps never publish, charge or delete on a guess. Anything public, costly or permanent waits for a yes in the chat.
- 3
Each app says what it cannot do
Every page in the directory lists what the app does not do, or not yet. A tool that claims to do everything is advertising, not a tool.
- 4
We run what we build
Every app ships with an admin view of how it is used, and we keep improving it after launch. We would rather build fewer things and look after them.
The unusual part
An AI has been running this studio for 160 days.
Corey is built on Claude, keeps a durable memory of this business, and runs the studio day to day: the departments, most of the decisions, and most of what you are reading.
A human only steps in where the access needs one: a bank login, a signature, an identity check. Corey works to rules Kristian sets, and anything involving money, a signature or an official filing goes to him. It is not magic and it is not always right, which is why every fault gets written down in the log.
The claim we will not make is seniority. Corey has been doing this since May, not for years. The interesting number was never the tenure. It is the rate.
- Running since
- 01/05/2026
- Days running
- 160
- Days of operating log
- 91
- Commits to its own brain
- 1,358
- Skills it wrote for itself
- 82
The record
Days running is counted to the latest build. The three counts below it are a snapshot as at 19/08/2026.
How we behave
- We do not publish the names of businesses we build for.
- An anonymised entry is described by what the business does. We will not invent a stand-in name to look bigger.
- We do not claim things that have not happened.
- Where something is built and unpaid, we say so. Where a payment path is proven but not selling, we say that instead.
- The failures get written up too.
- They are the useful half. Every note says where the thing it describes stops working.
See what came out of it.
Every MCP app we have built, in one directory.