I work full-time just like everyone else on earth, and at some point I think I may have thought it was not challenging enough so I do master’s studies on the side on top of that. And of course, I have a personal life to manage.
For months, I built what I thought was the perfect Obsidian vault. PARA methodology. Zettelkasten connections. Numbered folder prefixes. Twenty-five templates. Separate folders for Areas, Resources, Archives. Instruction files for AI assistants that read like technical documentation.
It was beautiful. It was organised. And it was completely unusable. I was personally having problems even finding the smallest things on my own. This was not a second brain. It was a second burden.
The Over-Engineering Phase
I started with Tiago Forte’s PARA method : Projects, Areas, Resources, Archive. The idea is elegant; organise by actionability, not by topic. Active work in Projects, ongoing responsibilities in Areas, reference material in Resources, completed stuff in Archive.
Then I layered Zettelkasten on top. Atomic notes. Permanent notes. Literature notes. Maps of Content. Every idea deserves its own note, linked to related concepts, building an interconnected web of knowledge.
My folder structure looked like this:
Personal/
├── 0-Inbox/
├── 1-Projects/
├── 2-Areas/
│ ├── Health/
│ ├── Finance/
│ ├── Relationships/
│ └── Goals/
├── 3-Resources/
├── 4-Archive/
├── Templates/
└── Attachments/
I had 25 templates across three “pillars” (Personal, Work, Study). Daily Note templates. Weekly Review templates. Literature Note templates with citation fields. Meeting Note templates with action item sections. Project templates with milestone trackers.
I even wrote instruction files for AI assistants: CLAUDE.md with detailed rules about which template to use when, which folder things belong in, what tags to apply. Over 150 lines of instructions per pillar.
It was a productivity system designed by someone who spent more time designing productivity systems than actually being productive.
What Went Wrong
The first sign of trouble: I stopped taking notes.
A friend recommended a book. I opened Obsidian to jot it down. Five minutes later, I was still deciding: Is this a Literature Note or a Zettelkasten permanent note? Does it go in Resources or Areas? Should I tag it #reading or #recommendations? Do I need the full Literature Note template with citation fields, or just a quick note?
I closed Obsidian and texted myself the book title instead.
That friction adds up. A system meant to reduce cognitive load was creating it.
The numbered prefixes were weird. Why am I looking at “0-Inbox” instead of just “Inbox”? The numbers existed to force alphabetical sorting, but they made everything feel like a database schema rather than a place for human thoughts.
Areas vs Resources was confusing. Is “Health” an area or a resource? What about notes that don’t fit either? Every time I created a note, I was playing a classification game instead of writing.
Archive was useless. I never archived anything. Old projects just sat in Projects until I deleted them. The Archive folder existed in theory but not in practice.
Templates became bureaucracy. Most of my notes didn’t fit neatly into any template. I’d use a template, delete half the sections, and end up with something that could have been a blank note.
The AI instructions were documentation, not action. Claude and Copilot would follow my 150-line instruction files, but the instructions were so detailed that getting AI assistance felt slower than just doing things manually.
The system was serving itself, not me.
The Insight
Look at how Anthropic and OpenAI designed their Projects features. The product design is remarkably simple: create a project, attach some files, start chatting. The project becomes your context. No configuration. No methodology. No decisions beyond “what should I call this?”
That’s brilliant product design. Zero friction to value.
The same teams that build the most sophisticated AI models in the world chose the simplest possible UX for organising work: a project is just a folder with stuff in it.
Not a folder following PARA principles. Not a Zettelkasten container with atomic notes and bidirectional links. Just a folder with the things you need for that project. The folder IS the context.
If that’s good enough for Claude and ChatGPT, it’s good enough for my Obsidian vault.
The Simplification
I gutted everything. Here’s what my Obsidian vault looks like now:
Personal/
├── Inbox/ # Quick captures
├── Projects/ # Active work
└── Knowledge/ # Everything else
Three folders. That’s it.
Inbox is for quick thoughts. No template, no tags required. Just capture and move on. I process it when I have time.
Projects is for active work with a deadline. A project folder contains everything related: notes, PDFs, images, whatever. It’s a self-contained workspace, like a ChatGPT project.
Knowledge is for everything else. Reference material. Ideas. Concepts. Learnings. If it doesn’t have a deadline and isn’t a quick capture, it goes here.
I deleted all 25 templates. If I need structure, I’ll create it in the moment. Most notes don’t need structure - they need words.
I removed the numbered prefixes. “Inbox” instead of “0-Inbox”. Names that make sense without mental translation.
I killed Archive. When a project is done, I archive it to Knowledge with a summary and delete the original. The commit hash lets me restore if needed.
Why Obsidian? Because I’m not locked in. My notes are plain markdown files in folders. If Obsidian disappears tomorrow, I open the same files in VS Code or any text editor. iCloud syncs them to my phone. Git tracks every change. And crucially: any AI assistant can read and edit plain text files directly. No API, no export, no integration needed.
I considered just using Claude Projects for everything. The UX is beautiful. But my knowledge isn’t static. I update insurance notes when claims change. I add to my 1:1 feedback after every meeting. I revise goals quarterly. Claude Projects are snapshots. My knowledge is a flowing river that needs constant updates, version history, and offline access.
Making It AI-Native
The simplification wasn’t just about fewer folders. It was about acknowledging an uncomfortable truth: AI assistants are already better than me at finding and summarising my own notes.
The first time I asked Claude “what did I discuss with Sarah last month?”, it scanned my 1:1 notes and answered in seconds. When I needed my dental insurance limits, it found the policy document I’d uploaded months ago and forgotten about. Delegating retrieval to AI isn’t laziness. It’s acknowledging they’re genuinely better at it.
So I built a system that leans into this. I capture and organise. AI retrieves and synthesises.
The manifest system. Each Knowledge/ folder has a MANIFEST.md file - a simple table listing every file with a one-line description. When Claude or Copilot opens my vault, they can read the manifest to instantly understand what’s there without scanning everything.
# Knowledge Manifest
| File | Description |
|------|-------------|
| [[2025 Goals]] | Personal goals and objectives for 2025 |
| [[API Design Principles]] | Notes on REST API best practices |
Multiple instruction formats. Different AI tools read different files:
CLAUDE.md- Detailed instructions for Claude CodeAGENTS.md- Universal format that works with any AI coding agent.github/copilot-instructions.md- Quick reference for GitHub Copilot
You don’t need all three. If you only use Claude Code, keep CLAUDE.md and delete the rest. If you use multiple AI tools (Copilot at work, Claude for personal projects), keep all of them. The content is similar - just pick what matches your tools.
Minimal decision tree. The instruction files went from 150+ lines to about 50. Here’s all an AI needs to know:
Quick thought? → Inbox/
Active work with deadline? → Projects/
Everything else? → Knowledge/
Simple rules, not complex documentation.
The Automation Layer
The final piece: automation that disappears.
Manifests update themselves. A Claude Code hook triggers whenever I edit files in Knowledge/, or a GitHub Action syncs on push. Either way, the index regenerates automatically. I never think about it.
Projects archive themselves. When I say “archive the project”, the AI creates a summary in Knowledge/, stores the commit hash for restoration, and deletes the original. The knowledge is preserved; the clutter is gone.
Weekly cleaning runs itself. A GitHub Action checks every Monday: Are manifests in sync? Items stuck in Inbox? Stale projects? If it finds anything, it creates an issue I can assign to Copilot, Claude, or handle manually.
The philosophy: automation should be invisible. Not another feature to configure. It just happens.
See the README for implementation details.
What Actually Matters
After simplifying, I discovered what makes a note-taking system work:
Capture without friction. The path from thought to saved note must be instant. Any decision point - which folder, which template, which tags - adds friction that kills capture.
Projects as workspaces. Anthropic and OpenAI got this right: the project folder IS the context. Put the PDFs in there. Put the meeting notes in there. Put the code snippets in there. Self-contained, like Claude Projects.
Knowledge should be discoverable. The manifest system means AI assistants can navigate my knowledge without me explaining it every time. They read the index and understand what’s available.
Tags over folders. I still use tags (#context/personal, #type/meeting, etc.) but they’re optional, not required. A note can exist without tags. Tags add findability without adding friction.
Automation should disappear. The best automation is the kind you forget exists. Manifests update themselves. Projects archive themselves. The system maintains itself while I focus on the actual work.
The Lesson
Productivity systems are seductive because they feel like progress. Building the perfect PARA + Zettelkasten hybrid felt productive. Creating 25 templates felt productive. Writing detailed AI instruction files felt productive.
But productivity theatre isn’t productivity. The best system is the one simple enough that you forget it’s there.
If you’re building your own second brain:
- Start with three folders (Inbox, Projects, Knowledge)
- Add a manifest for AI discoverability
- Automate the maintenance
- Let AI handle the busywork
- Add complexity only when you feel genuine pain
You probably won’t need to add complexity. I didn’t.
The goal isn’t a beautiful system. It’s a quiet mind.
P.S. If you want to try this approach, I’ve packaged it as a template: minimal-second-brain . Click “Use this template”, clone it, open in Obsidian. The automation just works.