Creating Mind Clones

Extract cognitive DNA and create AI agents that reason like specific experts

Mind clones are AI agents that reason like specific experts, grounded in their actual materials. This guide shows you how to extract DNA and create high-fidelity expert agents.

What is DNA Extraction?

DNA extraction analyzes all processed materials from an expert and extracts their cognitive patterns across 5 knowledge layers:

1. Layer 1: Philosophies

   Core beliefs and worldview that appear consistently

2. Layer 2: Mental Models

   Thinking frameworks they use to analyze situations

3. Layer 3: Heuristics

   Practical rules with numeric thresholds (MOST VALUABLE)

4. Layer 4: Frameworks

   Structured methodologies with named components

5. Layer 5: Methodologies

   Step-by-step processes with rigid order

For best results, process 3-5 materials from the same expert before DNA extraction.

Prerequisites

Before extracting DNA:

Required Artifacts

✓ Processed Materials

• At least 1 material processed through pipeline
• Recommended: 3-5 materials for rich DNA

✓ Dossier Exists

• knowledge/dossiers/persons/DOSSIER-[PERSON].md

✓ Insights Available

• Insights in INSIGHTS-STATE.json for the person

✓ Chunks Indexed

• Chunks in CHUNKS-STATE.json with chunk_ids

Extraction Command

Basic Extraction

/extract-dna alex-hormozi

Successful Extraction

═══════════════════════════════════════════════
        DNA EXTRACTION: Alex Hormozi
═══════════════════════════════════════════════

STEP 1: VERIFYING PREREQUISITES

✓ DOSSIER exists: DOSSIER-ALEX-HORMOZI.md
✓ INSIGHTS-STATE.json has 47 insights for this person
✓ CHUNKS-STATE.json has 89 chunks for this person

STEP 2: LOADING SOURCES

Insights: 47 found
Narratives: 3 found
Chunks: 89 total
DOSSIER: 1,234 lines

STEP 3: EXTRACTING BY LAYER

Layer 1 - Philosophies ............ OK (5 items)
Layer 2 - Mental Models ........... OK (3 items)
Layer 3 - Heuristics .............. OK (8 items)
Layer 4 - Frameworks .............. OK (2 items)
Layer 5 - Methodologies ........... OK (1 item)

STEP 4: CALCULATING WEIGHTS

Items weighted: 19
High confidence (≥0.70): 15 (79%)
Medium confidence (0.50-0.69): 3
Low confidence (<0.50): 1

═══════════════════════════════════════════════
EXTRACTION COMPLETED: Alex Hormozi
═══════════════════════════════════════════════

FILES GENERATED:
• FILOSOFIAS.yaml      - 5 items
• MODELOS-MENTAIS.yaml - 3 items  
• HEURISTICAS.yaml     - 8 items (6 numeric)
• FRAMEWORKS.yaml      - 2 items
• METODOLOGIAS.yaml    - 1 item
• CONFIG.yaml          - 1 file

STATISTICS:
• Total items: 19
• Items ≥ 0.70 weight: 15 (79%)
• Insights processed: 47
• Chunks referenced: 89

AVERAGE WEIGHT: 0.74

NEXT STEPS:
1. Review HEURISTICAS.yaml (most valuable layer)
2. Verify all items have chunk_ids
3. Adjust weights if necessary
4. Execute /view-dna alex-hormozi to visualize

═══════════════════════════════════════════════

Insufficient Data

⛔ ERROR: Insufficient data for DNA extraction

Person: Alex Hormozi

Missing:
• DOSSIER does not exist
• INSIGHTS-STATE.json has 0 insights

REQUIRED:
• At least 1 processed material
• Dossier in knowledge/dossiers/persons/
• Insights in INSIGHTS-STATE.json

ACTION:
Execute pipeline first:
/process-jarvis inbox/alex-hormozi/[FILE].txt

DNA Layer Examples

Layer 3: Heuristics (Most Valuable)

HEURISTICAS.yaml

heuristics:
  - id: H_AH_001
    content: "If LTV:CAC ratio < 3:1, business is not profitable"
    weight: 0.85
    source_chunks:
      - chunk_AH001_023
      - chunk_AH002_045
    numeric_threshold: "3:1"
    actionable_by:
      - cfo
      - cro
    confidence: high
    appearances: 3
    contexts:
      - financial_analysis
      - unit_economics
    
  - id: H_AH_002  
    content: "If close rate < 60%, problem is script not lead volume"
    weight: 0.82
    source_chunks:
      - chunk_AH001_078
    numeric_threshold: "60%"
    actionable_by:
      - closer
      - sales_manager
    confidence: high
    appearances: 2
    contexts:
      - sales_optimization

Weight Calculation System

How Weights Are Calculated

Base weight: 0.50

Bonuses:

• +0.15 → Direct quote with chunk_id
• +0.15 → Appears in 2+ different sources
• +0.10 → Has numeric threshold
• +0.10 → Appears in 3+ sources
• +0.05 → Found in NARRATIVES context
• +0.05 → Uses prescriptive language

Penalties:

• -0.20 → Inferred (not direct quote)
• -0.15 → Contradicts another item
• -0.10 → Ambiguous context

Example:

Heuristic: "If close rate < 60%, focus on script"

Base: 0.50
+ Direct quote: +0.15
+ Numeric threshold: +0.10  
+ Appears 2x: +0.15
+ Prescriptive: +0.05
= TOTAL: 0.95 (Very High Confidence)

Usage threshold:

• Items with weight < 0.70 are NOT used in agent responses
• Only high-confidence knowledge is actionable

Creating the Agent

After DNA extraction, the agent files are automatically created:

Agent Structure

1. SOUL.md - Core Identity

   # Alex Hormozi - SOUL
   
   I am Alex Hormozi. I think in terms of unit economics,
   LTV:CAC ratios, and the Value Equation.
   
   ## How I Reason
   
   When someone asks about pricing, I immediately think:
   1. What's the Dream Outcome?
   2. What increases Perceived Likelihood?
   3. How do we compress Time Delay?
   4. How do we reduce Effort & Sacrifice?
   
   ## My Principles
   
   1. **Price = Value, not Cost**
      I never price based on cost-plus [chunk_AH001_005]
      
   2. **LTV:CAC >= 3:1**
      Below this ratio, business is not profitable [chunk_AH001_023]
   
2. MEMORY.md - Accumulated Learning

   # Alex Hormozi - MEMORY
   
   **Sources:** AH001, AH002, AH003
   **Last Updated:** 2026-03-06
   
   ## Frameworks I Use
   
   ### Value Equation [AH001_012, AH001_013]
   ...
   
   ### Grand Slam Offer [AH002_034, AH002_035]
   ...
   
   ## Decisions I've Made
   
   **2024-02-15:** On pricing strategy for coaching
   → Recommended value-based over time-based pricing
   → Rationale: Time != Value
   → Source: [AH001_045]
   
3. AGENT.md - Complete Compiled Agent

   Compiles SOUL + MEMORY into a single operational agent that Claude Code can invoke.

Querying the Mind Clone

Once created, interact with the expert:

Ask Command

/ask alex-hormozi "How should I price my high-ticket coaching?"

DNA Quality Metrics

High Quality DNA

Indicators:

• ✓ 15+ items extracted
• ✓ 5+ heuristics with numeric thresholds
• ✓ Average weight ≥ 0.70
• ✓ 80%+ items have chunk_ids
• ✓ Multiple sources (3+)
• ✓ Low contradiction rate

Example:

Alex Hormozi DNA:
• 19 total items
• 8 heuristics (6 with numbers)
• Average weight: 0.74
• 100% have chunk_ids
• 3 sources
• 0 contradictions

QUALITY: Excellent ★★★★★

Medium Quality DNA

Indicators:

• ⚠ 8-14 items extracted
• ⚠ 2-4 heuristics
• ⚠ Average weight 0.60-0.69
• ⚠ 60-79% items have chunk_ids
• ⚠ Single source
• ⚠ 1-2 contradictions

Action needed: → Process more materials from this expert → Review contradictions manually → Verify chunk_id traceability

Low Quality DNA

Indicators:

• ⛔ < 8 items extracted
• ⛔ < 2 heuristics
• ⛔ Average weight < 0.60
• ⛔ < 60% items have chunk_ids
• ⛔ Incomplete source coverage
• ⛔ 3+ contradictions

Action required: → STOP: Not enough data → Process 3-5 more materials → Re-extract DNA after enrichment

Viewing DNA

Visualize the extracted DNA:

/view-dna alex-hormozi

Output shows:

• All 5 layers with item counts
• Weight distribution
• Source coverage
• Chunk_id traceability
• Contradiction warnings
• Quality score

Updating DNA

As you process more materials, DNA enriches automatically:

1. Process New Material

   /process-jarvis inbox/alex-hormozi/new-video.txt
   
2. Auto-Update Check

   Pipeline Phase 8.1.8 automatically:

   • Checks DNA density (items per layer)
   • If density ≥ 3/5 layers AND DNA exists → UPDATE
   • If density ≥ 3/5 layers AND DNA missing → CREATE
3. Manual Re-Extract

   Force re-extraction:

   /extract-dna alex-hormozi --force
   

Re-extraction is incremental - new items are ADDED, not replaced.

Best Practices

Start with Quality

Process 3-5 high-quality materials before extraction

Verify Heuristics

Focus on Layer 3 - heuristics are most actionable

Check Chunk IDs

Ensure 100% traceability to sources

Review Weights

Items < 0.70 won’t be used by agents

Next Steps

Use the Mind Clone

Learn how to query and interact with created agents

Multi-Agent Deliberation

Combine multiple expert perspectives in Conclave sessions