Whitepaper · AI factories
Building AI factories that deliver
Writing software got cheap. Running it inside a company did not. This paper describes an AI factory that turns requests into internal tools people use, and the foundation, floor and process it needs to ship them.
What it covers
A factory is judged by what it ships.
Coding agents made software cheap to write, so many companies now plan to replace bought tools with their own. Most of what a bought tool gives you is not its screens but what sits under them: data, permissions, hosting, upgrades, an audit trail. An AI factory delivers when those are built once, as a foundation, and every new tool is a small change on top that a person approves.
- Layer 01
Your teams
The people who use the tools
- Layer 02
Use cases
Quality, support, compliance
- Layer 03
Curiosity Studio
How tools get built, and approved
- Layer 04
Data
70+ systems, one graph
- Layer 05
Infrastructure
Your servers, with Orchestrator
Chapters
Seven chapters, from cheap code to the first tool
01 · Page 03Chapter 01
Writing software got cheap
What coding agents changed, and what they did not.
02 · Page 06Chapter 02
A line, not a lab
The six stations a request passes through.
03 · Page 09Chapter 03
The foundation decides what ships
What every tool inherits from underneath.
04 · Page 13Chapters 04 · 05
The floor and the way in
Orchestrator runs every Studio; Sudo and curio bring the work.
05 · Page 19Chapter 06
Features per day
What our own changelog measured.
06 · Page 22Chapter 07
Start with one tool
How to pick it, and what to ask.
Your data. Your infrastructure.