The Living Knowledge Vault and the Mini-Me Article

The opening

Lou opened with what he had been working on for the last several weeks. A system he was calling the Living Knowledge Vault, an extension of the Sukana engine work he had been teasing in earlier sessions. The thing he kept circling back to was the writing problem he had never been able to solve.

The pattern he described, the one most of the cohort recognized, was this. You learn something. You research it. You tinker with it. You write a couple of pieces. You teach it. The teaching part takes longer than the learning. Building a team of editors and writers to compress that work was never something he wanted to do. So the teaching backlog grew.

What changed

The vault now does the compression. He works in the moment, has the conversation, runs the experiment, takes the notes, and then a process at the end of the chat extracts the substance into the vault. The vault is structured as a knowledge graph, so when he later asks for an article on a related topic, the system pulls in not just the immediate context but related experiences, analogies, prior reasoning.

The breakthrough he wanted the cohort to feel was about voice. The articles that come out now sound like him. Not just structurally, the cadence, the asides, the “the other day I was doing this and noticed that” moves that he had never been able to coax out of an AI. The system has access to the actual asides, because he put them in the vault, so the model can reference real moments instead of fabricating plausible ones.

The mini-me framing

Lou described it as having a mini-me that goes through both the structural and experiential layer of the writing. The structure is the easy part. The experiential layer, the memory of having done the thing, was what was always missing. The vault is, in his words, a memory of experience, which he treats as an extra dimension to ordinary memory.

Mazie’s parallel

Mazie jumped in with the project she had given herself for the same week. Tax season. She had decided to learn how to take all of her receipts, automatically categorize them, and produce the summaries her accountant needed. She wanted to never have to manually do tax prep again. Lou shared the parallel version he had run, dropped all his account statements into a folder, told Claude to read every transaction, organize them by tax category, produce a detailed list and a summary. The work that usually took him a day or two took an afternoon.

What the cohort took away

The frame Lou closed with, the leverage shift is not from doing the work to delegating the work, it is from doing the work to capturing the residue of doing the work in a way that the AI can reuse on your behalf later. The vault is a residue capture system. The articles are the downstream artifact.

He offered to put the current version up on GitHub but suggested the better experience would be to walk the cohort through building their own vaults. The room agreed. That became the next two weeks of work.