Video summary

This AI System Will Make You So Smart It’s Almost Unfair

Main summary

Key takeaways

Technology

Technological concept / product idea

  • Build a fully AI-powered “brain” that provides persistent memory and context across everyday tasks (e.g., running companies, saving time, acting like the user).
  • Use an external knowledge base (files) so the AI doesn’t “forget” when a chat session ends.
  • Focus on Obsidian (Markdown + graph view) as the durable storage layer and context engine.

Step-by-step guide / tutorial flow

1) Go pro (paid AI tools)

  1. Recommend buying a pro/premium AI membership to access newer, faster models and better performance than free tiers.
  2. Claim: free versions use older/slower models; paid membership increases quality and priority.

2) Install the “brain” (persistent memory via files)

  • Key idea: most AI tools forget after the conversation closes; persistent memory must come from your own storage.

Storage options discussed

  • Google Drive
    • Works for files but is limited and “folder-like.”
  • Notion
    • Positioned as more AI-native due to richer context (not just folders).
  • Obsidian (preferred)
    • Uses Markdown files that are readable, visible, and linkable.

Obsidian benefits emphasized

  • Acts like a local wiki (on your computer).
  • Graph view makes it feel like a brain via relationships/links.

3) Give the AI an identity (agent personality + operating principles)

  • Uses three Markdown files to turn a generic AI into something that behaves like you:

    • user.md: who the user is, role, communication style, preferred frameworks, guiding principles
    • soul.md: tone/values + direct instructions for speaking/behavior (e.g., avoid hedging words)
    • identity.md: “what the AI is” (persona role, e.g., coach/chief of staff/accountability partner)

  • Pro tip: don’t write these manually—prompt another AI to interview you and draft the files.

4) Wire the brain (folder structure / organization)

  • Without structure, the AI “drowns in noise” and hallucinates.
  • Claim: accuracy improves from ~60% to ~85% after structuring folders.

Required folder set (“seven folders”)

  1. People
  2. Projects
  3. Decisions
  4. Companies
  5. Meetings
  6. Daily (short daily recap: 3–5 lines)
  7. Knowledge (reusable insights, frameworks, quotes)

Optional “8th” folder

  • MOC (Maps of Content): consolidated summary/index files linking many topic-specific notes
    • Example: a youtube.md MOC linking related frameworks across folders.

5) Feed the brain (extract and store the right info)

  • Use connectors to ingest raw sources (e.g., notes/recordings) into the Obsidian folder structure.
  • The AI should extract people/decisions/knowledge rather than storing useless raw data.

Meeting workflow example

  • Use Granola to auto-transcribe meetings.

  • In Granola settings, provide a custom prompt to extract:

    • Decisions (what, by whom, why)
    • Commitments (who promised what, by when)
    • Preferences (how people work/communicate)
    • Key insights (frameworks, strategic shifts, non-obvious observations)

  • Output is saved as Markdown into the Meetings folder, named like: YYYY-MM-DD meeting name.md

6) Compound the brain (overnight self-improvement)

  • The “brain” should periodically process, consolidate, summarize, connect, and prune.

Two refinement modes

  • Manual
    • Run a prompt to:
      • find orphan notes
      • consolidate duplicates
      • update MOCs
      • flag important items for review

  • Automated

    • Use Claude scheduled task + cron to run nightly (example given: 11:00)

  • Claim: each night the Obsidian graph evolves; pruning/cleaning improves context quality and therefore answer quality.

7) Use-case payoff (context-aware actions)

  • With the brain organized and linked, the agent can perform tasks like:

    • “Send the invite to John” by using the John node in the graph to retrieve contact details (email, cell)

  • Core claim: agents become more useful because they have retrieval-ready context instead of relying only on prompt text.

Reviews / analysis included

  • Compare storage tools (Google Drive vs Notion vs Obsidian) based on:

    • context capacity vs simple folder storage
    • readability (Markdown)
    • visualization (graph view)
    • how well the AI can use the structure

Key tools / products named

  • Obsidian
    • Markdown vault, graph view, local wiki
  • Claude
    • scheduled task / cron automation
  • Granola
    • meeting transcription + extraction into Markdown
  • Mentions connectors and an “AI extract” workflow (implied integrations)

Main speakers / sources (as stated)

  • Primary speaker
    • The creator of the system (first-person narrator; repeatedly says “I built…”, “I use…”, “I built a playbook…”, “find me on Instagram and DM…”).
  • Named third-party AI used
    • Claude (for scheduled automation)
  • Named transcription/extraction tool
    • Granola

Original video