I built this for myself first. My own company runs on an AI-native operating model: a git-owned brain holding every decision, constraint, goal, and piece of accumulated judgment, with AI agents that draft, check, and ship real work against it every day. Written specs the agents read at runtime. Review gates with calibration evidence. A weekly ritual that keeps the whole thing true.
It started as a productivity experiment and became the way the business operates. The uncomfortable discovery along the way: the models were never the constraint. The constraint was that nothing I knew was written down in a form an agent could trust, and once it was, the second constraint appeared immediately: nobody was checking the checkers.
A European industrial group brought me in to help their teams do the same thing, and the method turned out to transfer: extract the judgment from senior people and legacy code, structure it so retrieval is precise instead of probabilistic, govern it so it stays trustworthy as the business moves. Their internal teams tested the architecture against embeddings-based retrieval on their own data and converged on it independently.
That is now the practice: a fixed-price audit, a build sprint, and a governance retainer, for companies that want to own their brain rather than rent a knowledge layer from a platform.
Every company will have a brain; the question is who owns it. Microsoft, Google, and Glean will rent you a good one inside their walls. A brain you own is plain files in a git repository you control: portable across models, inspectable claim by claim, private to your infrastructure, and compounding in value with every decision it captures.
And whoever builds it, someone has to verify the verifiers. As teams delegate more checking to AI, the gates that decide what ships need their own evidence, their own audits, and their own escalation rules. That discipline is the part nobody does yet, and it is the part I refuse to skip.
Engagements run through Xanadu Labs LLC. I take a small number of clients at a time, because extraction is senior work and I do it personally. The Skill Engineer tool on this site is a standalone product from the same workshop. The blog is where the method gets written up as it develops.
Book a fit call