In this section I review one AI-powered application and demonstrate how it can be used to create new value.
For the past few weeks I have been building a personal knowledge base for myself.
The tool is gbrain, which Garry Tan, the CEO of Y Combinator, has been building in the open. It is a large set of markdown pages, plus an agent that knows how to search through them. Tan calls that combination the library plus the librarian. Mine runs locally against a Postgres database, it mirrors itself to a private repository every night, and I can reach it from my phone.
If the idea is new to you, I wrote a short primer on what a second brain is and how personal context management differs from ordinary note-taking. This is that idea with an agent sitting on top of it.
What I have been putting into it
What I have been doing since then is feeding it everything about me. I have been writing up the concepts I use and how I actually think about them. I have added the people and the organizations I work with, and what I really do with each of them. My work and my family are in there, and so are the things I have built and the ones I stopped building.
The raw material is whatever I already had: years of sent email, my CV, a LinkedIn export, podcast transcripts, meeting notes. It is up to about a thousand pages now.
It is more like onboarding an assistant than installing software
So what has this actually been like? Much more like training a new personal assistant, a human one, than I expected. It was difficult in the same places.
If you want that assistant to be truly helpful, you have to give them all the context. Much of that context turns out to be things you never stopped to articulate. Then they come back with questions, and some of those questions do not have a ready answer, so you have to work one out. They get things wrong and you correct them. They spend effort on something that does not matter and you tell them what to focus on instead.
None of it is technically difficult. It mostly just takes time and attention to detail.
That correction loop sits close to what I have described as AI memory, except that here the memory is mine to edit rather than something a vendor manages on my behalf.
The failure mode to avoid
Tan's own warning is that a brain treated as a dumping ground will give you a very confident agent that is wrong in ways nobody can trace. That is probably what you get if you skip the articulating and just pour sources into it.
What I have found useful against that is keeping a small set of hand-written pages that take precedence over everything else when the two disagree.
The part that applies to your organization
There is something a bit awkward about arguing for frontier teams and then going off to build a private brain for myself. It did teach me something about the team version, though.
Just a small part of my effort went into tools. It largely went into writing down how things actually work, who does what, and what I mean by the terms I use. I would expect roughly the same balance in a team that is working heavily with AI agents, and I suggest you plan for it.
If you want the ambitious version of this, Garry Tan's talk is worth the time. His own brain runs to something like 220,000 pages built up over 25 years, though he is careful to say that it started as a folder with a few markdown files about the companies he was working with and the people he kept emailing.
Your action step
Do not start with the tool. Open an empty document and write one page about a concept you use constantly at work and have never defined in writing. Pick something you would be mildly embarrassed to discover your team understands differently, for example what you mean when you call something a pilot.
Give that page to whichever assistant you already use, along with a real question that depends on it. The gap between the answer you get and the answer you wanted is the rest of the writing you owe it, and you can see that gap before committing to any particular tool.
You can find gbrain at github.com/garrytan/gbrain, and as always, reach out if you would like help thinking through where something like this fits in your own work.