You find a great article on pricing, bookmark it, then lose track of it and its insights within weeks. Sound familiar?
Or you have notes from a client conversation, a competitor’s website you wanted to revisit, a few ideas you jotted down on your phone, a screenshot from a webinar — all sitting in different places, disconnected, slowly becoming useless.
Andrej Karpathy, one of the most respected AI researchers in the world, a co-founder of OpenAI and former head of AI at Tesla, recently shared how he solves this problem for himself. His version is built around research papers and code, but the main idea works just as well for a service business, a marketing team, or anyone tired of losing their best ideas.
Before diving in, let’s break down the approach step by step, so you can see exactly how it works in practice.
The idea in one sentence
You create a simple folder on your computer to store everything: articles, notes, screenshots, and ideas. Then you let an AI organize it all into a connected, searchable knowledge base that you can actually ask questions to.
You don’t need a special app, a subscription, or a complicated setup. All you need are folders, text files, and an AI that organizes everything for you.
How it works
Step 1: Three folders
Create a folder on your computer called something like my-knowledge-base. Inside it, create three subfolders:
raw/ is your dumping ground for articles, notes, and anything you want to save.
wiki/ is where the AI writes the organized version of everything.
outputs/ is where answers to your questions get saved.
That’s the whole structure. You only need to use the raw/ folder. The AI handles everything else.
Step 2: Start dumping things in
Don’t overthink this part. Copy and paste articles into text files, save your notes, and drop in anything you’ve been meaning to read or reference. The main idea is that you aren’t organizing it; you’re just collecting it. Organization is the AI’s job.
Karpathy described it like this: raw source material goes in, and the AI “compiles” it into a wiki with summaries, connections between ideas, articles organized by topic, and an index that links everything together.
Step 3: Tell your AI to build the wiki
Using a tool like Claude, point it at your folder and give it a simple instruction: read everything in the raw folder and turn it into an organized wiki, with one file per topic and an index. Then you let it work.
When it’s done, you have something genuinely useful: a searchable, connected collection of everything you’ve gathered, organized by topic, with your AI already spotting connections you might not have noticed.
Step 4: Ask it questions
This is where it gets interesting. Once your knowledge base has enough in it, you can ask your AI real questions against your own collected material:
- “What do my notes say about handling difficult client conversations?”
- “Based on everything I’ve saved about pricing, what approaches haven’t I tried yet?”
- “Summarize the three main ideas across these articles I saved about email marketing.”
You’re not getting generic AI answers. Instead, you’re getting answers based on what you chose to save: your research, your notes, and your context.
Step 5: File the answers back in
When the AI gives you a useful answer or generates a report, save it back into the knowledge base. Karpathy calls this the compounding trick: each query builds on the base, making the next query more useful. Over time, your knowledge base gets smarter because it includes not only what you collected, but also what you’ve figured out along the way.
Step 6: Run a monthly health check
Once a month, ask your AI to review the whole wiki. Have it look for contradictions, find gaps, and suggest new topics you haven’t covered yet. This keeps your knowledge base accurate and growing in the right direction.
You might be wondering how this practical system fits real-world business. Here’s how it translates beyond AI research:
Here are a few ways this applies outside of AI research labs:
A marketing-focused business could build a knowledge base of competitor messaging, successful ad copy examples, customer feedback themes, and campaign notes. Then, you could ask it questions like, “What angles haven’t we tested in our email subject lines?”
A service business could collect client onboarding notes, common objection responses, pricing research, and industry articles. Then, you could ask, “What patterns show up across my best client relationships?”
A solo operator could use it as a genuine second brain for meeting notes, research, course materials, and ideas. This way, you can actually find and use things later instead of losing them to the bookmark graveyard.
You don’t need to be technical.
The setup Karpathy described is simpler than it sounds. To get started, make three folders and write a brief text file describing what your knowledge base should include and how you want it organized. Nick Spisak, who covers AI for business owners, points out that the AI only needs the folder structure and the instruction file to work—no specific app required.
If you can create folders and paste text, you can build this.
Ready to set this up for yourself? Here’s where to learn how:
This is exactly the kind of thing we cover in the Pro-How AI Club. We don’t just talk about the “what”—we show you, hands-on, how to do it for your specific business. Members get weekly AI updates like this one, plus live coaching and a community of small business owners figuring this out together.
Your notes, your research, and your ideas deserve a better home than a forgotten bookmark folder.
The best time to start was six months ago. The next best time is this weekend.
