Use AI to Build a Personal Knowledge Base

@karpathy dropped a post about using AI to build personal knowledge bases. 41K bookmarks. Thousands of replies debating tools, plugins, and setups.

Most people will bookmark it and never do anything.

Here’s what actually works. I’ve been running this system for months.

The Problem With Note Taking

Every note taking system I tried failed for the same reason.

I had to stop and organise. Decide where a thought lived. Pick a folder. Tag it. Structure it. That moment killed the habit every single time. Probably because I was overthinking it.

So I flipped the approach entirely.

Capture dumb. Organise smart.

The System

Four tools. That’s it.

• Obsidian for storage
• Wispr Flow for capture
• Claude Code for organisation
• Apple Reminders for actions

Every day I open a daily note in Obsidian. Throughout the day I use Wispr Flow to dump thoughts straight into it. Morning brain dumps. After meetings. When ideas hit. Raw thinking, unfiltered, out of my head and into one place.

I don’t organise anything. I don’t tag anything. I don’t decide where it goes.

That’s the AI’s job.

How Claude Code Handles Organisation

At the end of the day, Claude Code reads the daily note in full.

It understands the context. It identifies what matters. It takes key sections and categorises them into the folders I’ve set up. Work projects. Ideas. Education. Goals. Whatever structure makes sense for you.

If I mentioned following up with someone, it creates a reminder. If I flagged something to action next week, same thing.

I don’t touch any of it.

This is similar to what Karpathy described. A raw folder for unprocessed thinking. A wiki folder where the AI organises it. A simple schema file that tells the AI the rules.

No fancy plugins. No elaborate systems. Folders and a text file that acts as the instruction manual.

The Part That Surprised Me

A few months into using this system, something shifted.

I asked Claude to surface gaps in my thinking across a project I’d been working through. It came back with connections I hadn’t made myself. Ideas I’d mentioned weeks apart that were clearly pulling in the same direction.

I knew it somewhere. I just hadn’t looked at it directly.

This is the part most people miss when they set up these systems.

The value isn’t the organisation. The value is the compounding context.

The AI working with you after a few months of daily notes is fundamentally different from day one. It has seen how you think. What you care about. What keeps coming up. What you say you’ll do versus what you actually do.

That context changes everything.

The Setup Takes 30 Minutes

Create a folder structure. Raw inputs. Organised wiki. Outputs.

Write a simple schema file. Tell the AI what the folders are for, how to organise things, and what rules to follow.

Start capturing. Don’t overthink it. Just dump your thinking into one place every day.

Let the AI handle the rest.

Karpathy keeps his schema “super simple and flat.” Just a nested directory of markdown files. No database. No plugins. No elaborate tool stack.

The people who will benefit from this aren’t the ones debating Obsidian versus Notion versus Roam.

They’re the ones who set it up this weekend and start capturing on Monday.

Why This Actually Works

Most productivity systems fail because they add friction at the moment of capture.

This system removes it entirely. You just talk. You just dump. You just get thoughts out of your head without thinking about where they go.

The organisation happens later, automatically, by something smarter than a folder structure you designed once and forgot about.

And the longer you use it, the more valuable it becomes.

Not because you built a better system. Because you built a better context.

That’s where the productivity gain actually lives.

Stop debating tools. Start capturing.

The system is three folders and a schema file. The real work is showing up every day and dumping your thinking into it.

Everything else handles itself.