Making Knowledge Machine-Readable
Brian Madden has built a public knowledge system that lets AI draw on his thinking in real time. The same methodology sits at the heart of Praxis - and he deserves the credit.
Brian Madden has built something I wish existed when I started thinking about how AI should work with professional knowledge. His public knowledge repository at brianmadden.ai is a worked example of an idea that most organisations are still trying to articulate: if you want AI to work well with your thinking, your thinking has to be structured in a way that AI can actually work with.
This post is about what he has built, why the principles matter, and why it connects directly to what Praxis teaches.
What brianmadden.ai is - and what it is not
The FAQs on Brian's site are the best starting point and they are worth reading carefully. The distinctions he draws are precise, and each one rules out a different misunderstanding.
Is it a chatbot?
No. It is a data source your AI connects to. You talk to your AI, and it draws on Brian's knowledge when relevant. It is more like an always-updated context and knowledge source for your chatbot than a chatbot itself.
Is it a digital twin?
No. A digital twin simulates a person. This makes published thinking accessible. The distinction matters: a simulation tries to reproduce behaviour; a knowledge repository makes reasoning transparent and reusable.
Where does it come from?
Brian maintains a personal AI-powered knowledge system - sometimes called a second brain - which is how he uses AI on a daily basis. That system has everything he needs to work: content, ideas, his to-do list, thinking, meeting notes, documents, plans. brianmadden.ai is a subset of that private system, updated daily. Content flows from the private system to the public repository using explicit publishing principles.
How current is it?
The "current thinking" file updates frequently. The synthesis updates when new posts are published. Crucially, the AI flags stale content rather than presenting old thinking as current - a design decision that most knowledge systems do not make.
Can you fork it?
Yes. The GitHub repo is the source of truth. Fork it, build on it, use the frameworks in your own work.
Why these principles matter
What Brian has built is a demonstration of something important: the gap between "AI gives good answers" and "AI does my actual work reliably" is almost always a knowledge-structure problem, not a model problem.
A browser chatbot does not have access to your documents, your precedents, your institutional knowledge, or your professional judgement. It guesses what a reasonable answer looks like based on everything it was trained on. That is genuinely useful for one-off questions. It is a liability for anything you want to run consistently, at scale, with your specific standards applied.
The solution is not a better model. It is better-structured knowledge - explicit, current, machine-readable, and forkable.
Why Praxis teaches this methodology
Praxis is the training programme that teaches you to build that structure for your own work. The skill file - the central artefact in every Praxis session - is the working implementation of exactly this idea. Professional knowledge written down in plain English, structured so that a script can read it and apply it the same way, every time.
The Praxis course materials are themselves built on this principle. The course content is maintained as a structured knowledge repository - a folder of markdown files, organised by topic, written to be read by both humans and AI. You can explore it in the Praxis course library.
The credit
The idea of publishing a personal knowledge system as a GitHub repo - open, shareable, and forkable by anyone - is something Brian pioneered, and I want to attribute that to him directly.
If this framing is useful to you, his site is the place to go deeper.