TL;DR

I spent years meticulously building folder structures for my notes. First in OneNote, then in Obsidian. It felt productive and effective. It wasn't. The turning point was realizing that folders impose a single hierarchy on information that doesn't have one. What followed was a slow evolution: folders to tags, tags to relationships, relationships to semantic search, and finally to a system where I can ask my notes a question and get an answer with sources. This post is about that journey and what I learned at each stage.

The Folder Trap

I was a heavy OneNote user for years. It was perfect for me, at that time. I had nested sections inside sections inside notebooks, color-coded and carefully maintained. When I needed something, I navigated a hierarchy I'd built from memory. It worked as long as I remembered where I'd put things, and as long as the information fit neatly into one category.

It often failed to fit into just one category. This was frustrating, but I lived with it.

Does a note about an API authentication pattern live under "Security" or "Backend" or "Project Beta"? In a folder system, you pick one and hope you remember which one you picked six months later. Or you duplicate it. Or you create a shortcut. Every solution is a workaround for the fact that folders force a single axis of organization onto information that lives on many axes at once.

Folders are how most of us always organized things. On our computer, in our email, in our head. It felt like the natural way to think about information.

Tags, Properties, and the Web of Connections

When I moved to Obsidian, I carried the habit with me. My first vault looked like my OneNote notebooks: deep folder trees, carefully sorted. Obsidian was a better tool, but I was using it the same way. It was a filing cabinet with a nicer interface.

The shift started when I stumbled onto the idea of organizing notes with tags and front matter, or page, properties instead of folders. Obsidian supports a YAML block at the top of every note where you can define structured metadata: tags, relationships to other notes, sources, status. Of course, Obsidian puts that into a nice UI so you don't have to deal with raw config. Instead of putting a note inside a folder called "Security," you tag it knowledge/security and also knowledge/backend and also link it to Project Beta. The note exists in multiple contexts simultaneously without being duplicated.

More importantly, I started using relationship properties — up for parent notes, sibling for lateral connections, source for references. This created a web of explicit connections between notes that existed independently of where the files sat on disk. The physical location became secondary. The meaning was in the metadata.

I still manually tag every note. I'm an obsessive about it. But that habit, which felt like overhead at the time, turned out to be the foundation for everything that came after.

Breaking Out of the Vault

Once I'd internalized the idea that connections matter more than location, a new friction became visible: Obsidian's vault boundary. Obsidian's vault is this concept of using a high-level root folder to store all your Markdown files, the files for your notes. Inside this vault, everything was linked and discoverable. Outside it, nothing was connected. Not my code projects, not my office documents, not the reference PDFs. The vault was a better-organized island, but it was still an island.

This clicked during a conversation with a colleague. He'd tried Obsidian and bounced off it because his actual work didn't live inside the vault. That conversation led me to build the FolderTether Obsidian plugin. It creates a bidirectional link between a note and any directory on the filesystem. From Obsidian, one click opens the linked folder in Finder. From Finder, a .url file sitting in the directory opens the note directly in Obsidian. The vault stops being a walled garden and becomes a navigational index over your entire digital workspace.

It's a small tool, but it completed a shift in how I think about notes. The note about Project Beta doesn't need to live in a "Projects" folder. It doesn't even need to be near the project files. It just needs a connection to them. The same principle that moved me from folders to tags now extended beyond the vault itself: physical location is secondary to relationships.

And this turned out to matter for AI agents too. When Claude Code queries my vault, it can discover notes about projects and follow the linked_dir property to find the actual project directory without me having to tell it where things are. The vault became an index not just for me, but for any tool that can read Markdown and follow links.

When Good Organization Still Isn't Enough

Here's where most people stop, and honestly, where I stopped for a while too. Tags and links are a genuine improvement over folders. You can find things by topic, follow connections between ideas, and Obsidian's graph view gives you a visual map of how everything relates. For a personal knowledge management system, it's good. For hundreds of notes, it's great.

For thousands of notes, it breaks down in a different way.

The problem at "scale" isn't organization anymore, it's retrieval. You can tag perfectly and still not find what you need, because you're searching for a concept and your note used different words. Search for "authentication" and you won't find the note you wrote about "session token management" or "login security." Most people would expect that they're about the same thing. Keyword search finds string matches. It doesn't understand meaning.

This is where the journey took another turn. One that only happened after I started working with AI and implementing RAG (Retrieval Augmented Generation). Semantic search is based on vectors (numerical representations of meaning) and search is conducted based on similarity rather than exact text match. You ask "how do I handle user authentication?" and it returns notes about session tokens, OAuth flows, and login security, because it understands that these concepts are related even though they share no keywords.

Once you've experienced semantic search, keyword search feels like searching for a book by the color of its cover.

Seeing How Ideas Connect

But even semantic search has a ceiling. Although it finds relevant documents, it doesn't understand how they relate to each other. It can tell you that five notes are all relevant to "AI," but it can't tell you that two of them describe the same LLM provider, that one references a model that is no longer state of the art, or that three of them were written about the same agent framework.

That's what a knowledge graph adds. Instead of treating notes as isolated documents that happen to match a query, a knowledge graph extracts entities and the relationships between them. It builds a structure that mirrors how the information actually connects. You stop asking "which documents match?" and start asking "how are these ideas related?" and "which projects use this technology?" and "what else have I written about this tool?"

The graph doesn't replace search. It sits on top of it. Semantic search finds the relevant notes; the knowledge graph understands the connections between them.

Ask Your Notes a Question, Get an Answer

The reason I'm writing this series is to shine a light on what happens when you combine all of this with an AI agent like Claude Code.

The setup: Obsidian holds the notes. A semantic index makes them searchable by meaning. A knowledge graph maps the relationships between entities. And an AI sits at the top, receiving your question, retrieving the relevant context from all three layers, and generating an answer with citations back to your actual notes.

This is RAG applied to your own knowledge base. You ask a question in natural language. The system retrieves the relevant context from your notes. The model generates an answer grounded in what you've actually written, with links to the source notes so you can verify and go deeper.

It sounds like a lot of infrastructure for personal notes. And if you have fifty notes about recipes and travel plans, it is. But if you have years of accumulated knowledge across projects, research, decisions, and ideas, it changes what your notes are for. And as someone whose projects and ideas regularly cross the line between private and professional, having a way to connect and query all of it is a game changer.

I didn't plan this arc. I didn't sit down one day and decide to build a knowledge management pipeline. I started by moving my notes out of OneNote. Then I got frustrated with folders. Then I discovered tags. Then I hit the limits of keyword search. Each step solved a problem that the previous step had revealed. And each step, it turns out, was laying groundwork for the next.

I'm also lucky with timing. If I'd done this reorganization work two years from now instead of two years ago, the AI tools would exist but my notes wouldn't be ready for them. The tagging, the relationships, the structured note properties, all of that manual work I did before AI agents were part of my workflow is exactly what makes the notes consumable by AI agents now. The lesson is that investing in how you organize information pays off in ways you can't predict at the time.

What's Next

This is the first post in a nine-part series on building an AI-powered knowledge management system. The next post covers Obsidian specifically, not as a note-taking app, but as a knowledge platform, and why Markdown files on disk turn out to be the ideal foundation for everything described above.

This post is part of the AI-Powered Knowledge Management Series. Next: Obsidian as a Knowledge Platform.