The PARA Method in Obsidian with Claude: Why I Modified It

TL;DR

Tiago Forte's PARA method (Projects, Areas, Resources, Archive) is the most popular knowledge-management framework of the last five years. It works well for human note-taking, but applied straight to an AI second brain, it produces a vault that's noisy for the AI and overhead-heavy for you. The fix is small: keep Projects, drop or simplify Areas, treat Resources as canonical files only (not collections), and archive aggressively. This post covers what works, what breaks, and the PARA variant I use day-to-day with Claude + Obsidian.

What PARA Originally Solves

PARA was designed by Tiago Forte to organize digital information for actionability. Four categories:

The framework's strength is that it organizes around what you'll act on next, not what topic something belongs to. That's a meaningful improvement over folder hierarchies built around subject matter.

For humans browsing their own notes, PARA is excellent. The graph of "projects → areas → resources" maps onto how the human brain prioritizes attention.

Where PARA Breaks for AI

Applied straight to an AI second brain, three problems emerge:

1. Areas produce sprawl. "Health," "Finances," "Career" sound clean as labels, but the actual files inside accumulate fast: random observations, half-finished thoughts, articles you read once. The AI has no way to know which of these matter and which are noise. A 12-file "Career" area folder contains 2 useful files and 10 distractions, and the AI surfaces all of them.

2. Resources become a junkyard. "Resources" in PARA is everything that isn't a project or area. In practice this becomes a dumping ground for clipped articles, screenshots, and reference material you haven't read carefully. For human PKM, that's fine, you can re-encounter resources later. For AI, every file in Resources competes for context budget.

3. The framework rewards organizing over thinking. PARA's elegance encourages you to spend time placing notes correctly. That time is mostly wasted from the AI's perspective. The AI doesn't care whether a note is in "Areas/Health" or "Areas/Fitness", it cares whether the note has signal. Time spent organizing is time not spent curating.

The PARA Variant That Works

The fix isn't to abandon PARA. It's to weight the four categories differently for AI use:

Projects (heavy use). Keep this exactly as PARA designs it. One file per active project. The file is short: status, stack, recent decisions, blockers, key people. The AI uses this file constantly. This is the same as the 3-file template's projects.md.

Areas (use sparingly). Don't make folders for every life area. Have at most 3 - 5 area files, each one paragraph long. "My financial principles," "How I think about health," "Career direction." If an area file grows past 200 words, you're probably writing for yourself, not for the AI. Move the long-form content to a personal journal that the AI doesn't load.

Resources (canonical files only). Reframe "Resources" as "Reference files I want the AI to know about." Examples: glossary of technical terms specific to your industry, list of vendors and what they're for, opinions you've formed on common debates (Vite vs Next.js, Postgres vs MongoDB). Don't put clipped articles or random highlights here. If a resource isn't load-bearing for AI decisions, it doesn't belong.

Archive (aggressive). Anything that hasn't been actively useful in 60 days goes here. The AI doesn't load the archive folder. This is the most important PARA habit; without it, the vault decays. Schedule a quarterly archive review and be ruthless.

Mapping PARA → AI Vault Structure

A practical folder layout:

brain/
├── CLAUDE.md            # = identity.md, loaded automatically by Claude Code
├── projects.md          # PARA Projects: all active projects in one file
├── decisions.md         # NOT in original PARA: append-only decisions log
├── areas/
│   ├── financial.md     # one paragraph each
│   ├── health.md
│   └── career.md
├── resources/
│   ├── stack.md         # canonical opinions on tools
│   ├── glossary.md      # industry-specific terms
│   └── vendors.md       # who I use for what
└── archive/             # not AI-loaded; quarterly review
    ├── projects/
    ├── areas/
    └── resources/

Total active files: under 15. Total file content: under 5,000 words. Every file in the active set earns its place by being something the AI actually uses, not by feeling productive to write.

Why I Added "Decisions" to PARA

Tiago Forte's original PARA doesn't have a decisions file. I added one because, in practice, decisions are the single most useful thing for AI context.

A project file says what you're doing. An areas file says what you care about. A resources file says what you know. But none of them say what you've already decided and rejected. That's the gap.

decisions.md fills it. Every meaningful technical or operational choice goes in, append-only, with the reason and date. "We chose Vite over Next.js for SPAs because…" "We chose Supabase over Firebase because…" The AI uses this constantly to avoid suggesting paths you've already ruled out.

If you're going to add only one file beyond standard PARA, add this one.

Common PARA-in-AI Mistakes

Things I've seen people do wrong:

Mistake 1: One file per active project, but with too much detail. A project file should be one paragraph per project, not a sub-vault per project. If you find yourself making sub-folders inside Projects, you're back to noise.

Mistake 2: Areas folders mirroring life domains. "Family," "Hobbies," "Personal growth", these belong in a personal journal, not the AI second brain. Keep Areas tight to professional/operational domains.

Mistake 3: Resources used as a bookmarks folder. If you want to save articles, use Pocket or Raindrop. The AI doesn't need to read every article you've clipped. Resources should be your own synthesized opinions, not raw inputs.

Mistake 4: Skipping archive. This kills more second brains than any other single mistake. Without aggressive archive, the vault decays into a museum of how you used to think.

What Tiago Forte Got Right

Despite the AI-specific modifications, the core PARA insight (organize around actionability, not topic) is still correct. The variant above is PARA-adjacent, not anti-PARA.

If you've read Building a Second Brain and want to extend it for AI use, this variant is what I'd start with. If you haven't read it and just want an AI second brain, the 3-file template is the simpler starting point: PARA's structure is more than most people need.

The shared insight across PARA and the 3-file template: tight, actionable, curated structure beats sprawling personal-knowledge architecture for AI use. The exact folder names matter less than the discipline of curation.

brain/
├── CLAUDE.md            # = identity.md، يحمّله Claude Code تلقائياً
├── projects.md          # PARA Projects: كلّ المشاريع النشطة في ملفّ
├── decisions.md         # ليس في PARA الأصليّ: سجلّ قرارات للإضافة فقط
├── areas/
│   ├── financial.md     # فقرة لكلّ واحد
│   ├── health.md
│   └── career.md
├── resources/
│   ├── stack.md         # آراء مرجعيّة في الأدوات
│   ├── glossary.md      # مصطلحات خاصّة بالصناعة
│   └── vendors.md       # من أستخدم لماذا
└── archive/             # لا يحمّله الذكاء الاصطناعيّ؛ مراجعة فصليّة
    ├── projects/
    ├── areas/
    └── resources/