PKM Wiki

Source - YB 2026 - Claude-Obsidian Setup Tips

4 min read

Citation

YB (@yb_effect), X (Twitter) thread, 2026-04-13. URL: https://x.com/yb_effect/status/2043704637044211771. References YB's Substack "Engineering Agency" for detailed posts.

Overview

Practitioner thread describing a 9-month-old Claude + Obsidian workflow aimed at turning Obsidian into a personal thinking tool rather than a research wiki. The thread's central claim: the context you add personally — daily commonplace notes, atomic concept notes, highlights from intentional reading — is what makes the tool useful; you cannot outsource lived experience to AI. The thread contrasts this with Karpathy's LLM wiki base pattern, which YB characterizes as optimized for research projects and synthetically produced artifacts rather than personal thinking.

The Five Practices

YB lays out five sequenced practices:

  1. Daily commonplace practice. Jot down observations, epiphanies, thoughts, questions. Don't optimize for structure — optimize for getting it down. Use voice transcription because it's easier to say more, including hedges ("ifs, buts, ums") which carry signal about certainty. Context compounds over weeks.
  2. Atomic notes (Matuschak tradition). Think in concepts, not facts/names/dates. Treat notes as "API endpoints you call when trying to solve through problems in your professional or personal life." Write for yourself in a way that makes sense to you. Quotes @paulg: "writing is thinking." Eventually you reach the point where dictating a train of thought to Claude surfaces your own past thoughts to help solve current problems.
  3. Intentional reading habit. Connect Readwise or Obsidian Web Clipper. Audit past content consumption as "cope" (podcasts as productivity theatre, Substack scrolling). Cites @phokarlsson's "Structure your milieu" essay: ask who gets access to your brain, read them religiously, block everyone else. Scrolling = searching for reads to add to backlog; "consumption time" = highlighting and atomic-note creation. Even a 5-min Substack read should generate atomic notes and connections.
  4. Build an agentic constitution (.md). A portal between Obsidian and Claude. Contains: who you are, your big question, how you like to work, main goals, current problems, team context. Every agent action begins by reading the constitution so it acts in alignment. A constant WIP — YB updates it almost weekly. "Think about this doc like you're updating your team."
  5. Knowledge Work PRs. Subroutines (skills/agents) stored in the vault, each with specific instructions, all starting by reading the constitution. Example: "Tend the Vault" — reads daily commonplace entries since last trigger, notices threads, suggests atomic notes to update, drafts additions. YB then does a 5-minute voice session approving/tweaking/rejecting each change. The metaphor: agents do the work, user reviews the PRs like a manager, user approves to "production."

Key Takeaways (Practice Claims)

  • Context compounding is real if sustained: YB reports pattern-noticing after a few weeks of daily practice
  • Voice transcription > typing for daily commonplace capture (easier to say more, hedges carry meaning)
  • LLM wiki != personal thinking tool: YB distinguishes Karpathy's wiki pattern (research, synthetic artifacts) from personal-thinking use (accumulated lived context). Both are legitimate but different.
  • Skills as code methods: vault subroutines are analogous to methods in a code repo; the agentic constitution is the shared context; review loops are PRs to main
  • Audit consumption for cope: much "productive consumption" (podcasts, scrolling) is passive; intentional reading requires highlighting + atomic note generation per session

Concepts and Entities

Concepts

  • Daily commonplace practice (voice-transcribed)
  • Atomic notes as concepts, not facts (Matuschak tradition)
  • Atomic note as "API endpoint"
  • Intentional reading habit
  • Structure your milieu (cited from @phokarlsson)
  • Consumption-as-cope (anti-pattern)
  • Agentic constitution (.md)
  • Knowledge Work PRs
  • Subroutines / vault skills
  • "Tend the Vault" (example skill)
  • LLM wiki vs personal thinking tool distinction

Entities / Mentioned

  • YB (@yb_effect) — author, Substack "Engineering Agency"
  • Andy Matuschak (@andy_matuschak) — atomic notes practice
  • Andrej Karpathy (@karpathy) — LLM wiki base pattern
  • Paul Graham (@paulg) — "writing is thinking"
  • Phokarlsson (@phokarlsson) — "Structure your milieu"
  • Steph Ango / kepano (@kepano) — Obsidian Web Clipper
  • Readwise Reader
  • Claude (Anthropic)
  • Obsidian

Credibility Note

Tier: medium (practitioner report). Single-author X thread reporting 9 months of personal practice. No peer review; no controlled comparison; no quantified outcomes. Useful as a contemporary practice document (a 2026 snapshot of how practitioners are wiring Claude + Obsidian together) and as a source of concept vocabulary that may circulate further ("agentic constitution," "Knowledge Work PRs"). Not evidence for effectiveness claims. Treat claims like "context compounds" as hypothesis rather than demonstrated result.

Why This Matters for PKM

YB articulates a distinction the PKM community is still working through: the LLM-wiki pattern (Karpathy-style, synthetic, research-oriented) vs the personal-thinking-tool pattern (accumulated lived context, daily commonplace, atomic notes as personal concepts). Both are legitimate uses of Claude + Obsidian; they require different structures. YB's thread is one of the clearest short articulations of the personal-thinking-tool direction, and introduces two potentially useful contemporary concepts: agentic constitution as the LLM's orientation document, and Knowledge Work PRs as a review loop for agent-produced changes.