Challenge: every AI I opened started from scratch: it didn’t know my businesses, didn’t know how I decide, didn’t remember the previous conversation. I re-explained my own company at the start of every chat, because the context lived only in my head and in loose notes.
Solution: an external brain, outside any tool, that is the single source of truth about me and the businesses: memory organized by business, the way I actually decide, and agents that feed the base on their own. Any AI reads it first and arrives already knowing.
Results: I stopped starting over. Any model I open already works knowing the business and how I think, flags me when it senses I’m deciding in the wrong mode, and on a day of overload knows my context better than I do.
Every AI I opened started from scratch
I run several fronts at once, and I hit a ceiling early that most people don’t even see: each AI I opened didn’t know my businesses, didn’t know how I think, didn’t remember the previous conversation. I spent the start of every chat re-explaining my own company.
And there’s a second trap, less visible: the problem wasn’t the AI. It was that my context lived nowhere. It was scattered in my head, in loose conversations, in lost notes. No tool can know what was never written down. That isn’t a personal productivity problem, it’s structural.
I built the brain outside the tools
The obvious route would be building this inside the trendy product. But AI products change every week: they swap name, owner, price. If my context lived in there, every tool change would mean starting over. Tools die. Context survives.
So I built My Digital Cortex outside all of it: the single source of truth about me and the businesses, in a simple format any model can read, today and five years from now. It’s four layers, and each one only comes in when the one below is solid.
Memory, all the context in one place, organized by business. Persona, how I actually decide, not how I’d like to. Agency, agents that stop advising and start acting, like the one that takes care of my WhatsApp. And triggers, actions that fire on their own, without me asking.
Any AI arrives already knowing
In practice, it works like this: I open any AI and it reads the Cortex first. It arrives knowing the business and the way I think, and when it senses I’m in the wrong mode for an important decision, it flags me. An AI that only agrees is worthless; this one corrects me.
Each layer was born from a problem I hit the hard way: AI invents things to look useful, I don’t decide the way I think I do, and the automation that feeds the base makes mistakes. It took months of finding what breaks, and the principle holding it all together today is distrust.
Better to record nothing than to record something wrong.
Giving any AI permanent memory of the business, without tying it to any tool, is exactly the kind of project I help companies build.
Stack: simple text files version-controlled in Git and synced through Google Drive, opened in Obsidian; and a cascade of models (Claude, GPT, Gemini) that reads and feeds the base back, with the numbers computed in code and the AI only writing them up.