Do you love AI but keep having the nagging feeling that it’s not *quite* as powerful for you as it could be?
I finally built a system this week after realizing I’d had the same strategic conversation with AI, in different chat windows, at least four times. It was essentually different phrasing but the same conclusions, re-derived every time from scratch.
My thinking and decisions were getting lost as soon as I moved to another task.
And I found myself needing to upload the same base document to several chats, when I just wanted Claude to be able to pull information from across chats.
Enter: A second brain with an AI-maintained “LLM wiki” layer that turns your scattered notes, meetings, and AI conversations into one compounding source of truth.
The problem isn’t that leaders lack good ideas. It’s that the thinking behind those ideas isn’t being stored anywhere it can compound — so we pay the “figure this out” cost over and over, instead of paying it once and building on top of it.
If it’s happening to me, it’s happening to every leader using AI to think out loud and then losing the output.
I found a fix: a second brain with an AI agent layered on top — I call it my LLM wiki — that turns scattered notes and conversations into one compounding, cross-referenced source of truth. It takes about an afternoon to set up and will save you a lot of time down the road.
Quick summary:
- An LLM wiki is a knowledge base an AI agent maintains for you, turning scattered notes, etings, and AI conversations into one compounding source of truth.
- For leaders, that means decisions stop getting re-litigated, institutional knowledge stops living in one person’s head, and onboarding gets faster.
- It takes about an afternoon to set up: a raw folder, a wiki folder, and one schema file telling an AI agent what’s worth keeping.
But before we dive in, let’s think through a few other use cases first…
How can leaders use AI in Leadership?
Have you ever found yourself wondering why you made a decision? Why the team decided to do it “this way”…?
Or maybe you’ve had a conversation thinking through the ins, outs, and details, of a problem or a decision. The thinking you did with AI was gold! And you reached clarity, but now you can’t find it anywhere…
What about explaining things? Ever get tired of re-explaining things to different people on your team?
Or maybe you have someone on your team who you wonder, what will we ever do if they leave? HOW will the next person coming in possibly do the job with all the institutional knowledge stored in that employee’s brain?
My real-life frustrations essentially boil down to these questions:
- How can I use AI to remember thinking processes and logic around decisions (so I don’t have to keep re-making them)?
- How can I remember all the juicy nuggets I figure out with AI (instead of them getting lost)?
- How can I use AI to store my knowledge as a resource for my team?
- How can we capture the knowledge of a team member?
Imagine:
- Spending less of your time re-explaining,
- A team that can move without you in the room,
- Institutional knowledge preserved for good (without having to write SOPs!)
- Never losing track of an AI conversation or decision explanation (oh, and not having to re-upload critical files across multiple AI chats, while you’re at it!)
Enter: the wiki.
Aside from learning the ins and outs of human beings, it might be the single greatest leadership skill you ever acquire.
I know, that sounds dramatic, but seriously…
My experience with AI thus far has been 100x more powerful than my experiences with the advent of the internet itself.
And a wiki is an absolute game changer.
I finally decided to build one this week after realizing I’d had the same strategic conversation with AI, in different chat windows, at least four times. Different phrasing, same conclusion, re-derived every time from scratch.
My thinking and decisions were lost as soon as I closed the tab.
What a waste of time.
If it’s happening to me, it’s happening to every leader using AI to think out loud and then losing the output.
How do you make your thinking compound?
The problem isn’t that leaders lack good ideas,
It’s that the thinking and ideas aren’t being stored properly so that they can compound.
Every time you open a new chat with an AI tool and ask it to help you think through a decision, it starts from zero.
Yes, Claude has projects where you can upload data, but even then Claude doesn’t pull information across projects or into individual chats.
It doesn’t remember what you concluded last month about the reorg, and it doesn’t know you already ruled out that vendor for a reason. It re-derives the synthesis every single time – which means you pay the “figure this out” cost over and over, instead of paying it once and building on top of it. (And by “figure it out cost” I mean both your time and mental energy, as well as your AI usage, if that’s something you’re tracking.)
Multiply that cost across a leadership team.
Across a company.
Across the decisions that get re-litigated in every quarterly offsite because nobody wrote down why the last one was made.
That’s institutional memory leaking out in real time, and most of the time we don’t even notice it because each individual leak feels small.
What is an LLM wiki?
An LLM wiki is a knowledge base an AI agent maintains for you, turning scattered notes and conversations into one compounding, cross-referenced source of truth.
The first part is your knowledge base. Mine started years ago with a second brain, where I have been tucking away anything I found interesting, resources, and files inside a system of folders and notes in Obsidian.
Here’s what my second brain looked like as I started to set up the wiki:
he LLM wiki idea comes from Andrej Karpathy – AI researcher, former Director of AI at Tesla, and a founding member of OpenAI.
The AI consumes and interprets your content, actively organizes it, cross-links it, and builds wiki pages to summarize your content into themes. It has functions to maintain itself by periodically linting itself — catching broken links, orphaned pages, contradictions between pages, and claims that have gone stale, then flagging what needs a second look instead of quietly overwriting it.
The underlying logic applies directly to how leaders should be thinking about AI right now.
The structure is simple, and that’s the point:
Raw material stays untouched. Meeting transcripts, board notes, research, client history — none of it gets edited or reorganized. It just sits there as source of record.
A wiki layer sits on top. This is where an AI agent does the actual work: pulling the ten scattered mentions of a decision, a client concern, or a strategic bet across different sources and collapsing them into one page that keeps getting richer every time you add something new.
One schema file (Claude.md) tells the agent how to behave. What counts as worth a page. When to update versus create new. When to flag a contradiction instead of silently overwriting it.
The result? Well think of it this way.
A chatbot answers the question you ask right now.
A wiki already knows the answer, because the synthesis was done the first time the topic came up — and every time after that, your knowledge base got sharper instead of starting over.
One caution worth naming, because it’s the mistake I almost made: don’t build a wiki page for something that’s already small and easy to find in one place. It earns its keep on things scattered across many sources and many conversations — not on things that are already one clean file. Overbuilding this defeats the purpose of it.
How can leaders use AI in leadership, specifically?
Let’s talk leadership use cases:
- Decisions stop getting re-litigated. When the rationale behind a strategic call lives in a page instead of someone’s memory, the next person who questions it gets the reasoning, not a repeat of the debate.
- Institutional knowledge stops living in one head. The thing that makes a senior leader hard to replace is often just that they remember why. A wiki externalizes that, which is good for succession planning and terrible for nobody.
- Client and stakeholder history actually compounds. Instead of reconstructing context before every important conversation, the context is already there, already synthesized, already flagged for what’s changed since last time.
- Onboarding gets faster. A new hire — or a new leader stepping into a role — inherits the reasoning, not just the artifacts.
- Your own strategic thinking gets sharper. The version of you that’s had the same conversation with an AI four times in four different chats is not as sharp as the version of you building on a synthesis that’s already been through three rounds of refinement.
Why now, and not “eventually”
This tool is crazy simple. And if you have questions as you build it, guess what? Just ask AI! What used to require an internal wiki team now requires an afternoon of essentially vibe coding. It’s no longer an IT project. It’s a folder structure, one schema file, and a habit of telling the system “file this” before you close the conversation.
And the cost of waiting is compounding too — just in the wrong direction. Every quarter without this is another quarter of decisions made and forgotten, context rebuilt from scratch, and good thinking that never gets the chance to build on itself. The leaders who start now aren’t just early adopters of a tool. They’re the ones whose organizational memory starts compounding while everyone else’s keeps resetting to zero.
How to actually start (not a six-month rollout)
You don’t need a fully automated pipeline on day one. Here’s the simplest path in (and exactly what I did), stripped to the essentials:
- I started building a second brain in Obsidian a few years ago after watching:
Obsidian isn’t the only second brain tool, it’s just my favorite based on my needs and after doing my own research. Other tools include Notion, Roam Research, Logseq, and Tana. The tool matters less than the habit of putting everything in one place.
- I watched this video to get a basis of understanding on how to incorporate AI with my second brain:
- And then I watched this tutorial on how to actually build it:
- I pulled the Claude.md file from the description in the video above, fed it to Claude, and had it use what it knows about me to write the wiki’s purpose statement and add its own maintenance rules.
Here’s what my business wiki schema file looks like:
(and here’s a link to my actual file you can copy)
I know, this is a lot.
But remember, you don’t need a fully automated pipeline on day one. You can build slowly, just start!
1. Pick one thing that’s scattered across many sources — a recurring client concern, a strategic debate, a piece of institutional history nobody’s written down. Start there, not everywhere.
2. Separate raw from synthesized. Keep your source material — notes, transcripts, docs — exactly as it is. Build the wiki layer next to it, not instead of it.
3. Write down the one rule that matters most: what’s worth a page, what isn’t. This is the entire schema file to start. You can add more rules later.
4. End every important conversation the same way: tell the system what to keep before you close the tab. This is the single habit that makes the whole thing work, because there’s no clean way to retroactively pull decisions out of a chat history once it’s scrolled away.
5. Let it get messy before it gets clean. The first pages will be rough. That’s fine.
What’s Coming Up Next
This is the first piece in a new thread I’m building on AI-augmented leadership — the practical, not-hype version of what it actually looks like for leaders to use these tools well. Next up: a closer look at what’s worth turning into a wiki page and what isn’t (the overbuilding trap is real), and a piece on the one habit — “file this before you close the tab” — that determines whether any of this actually works.
If you’re already experimenting with AI tools as a leader and keep having the nagging sense that your best thinking isn’t going anywhere, this is the fix. It takes an afternoon to build and will have exponential benefits for you and your team.