Overview

The 4D Framework is Anthropic's model for effective AI delegation, organizing AI fluency into four interconnected components. Each dimension addresses a critical aspect of working with AI systems.

Core Principle: Effective AI use requires competence across all four dimensions—weakness in any dimension limits overall effectiveness.

When to Use This Skill

• Assessing overall AI fluency
• Structuring AI training programs
• Diagnosing AI effectiveness gaps
• Building systematic AI practices
• Teaching others effective AI use

The Four Dimensions

Dimension 1: Delegation

Question: What tasks should I give to AI?

Core competency: Selecting appropriate tasks for AI based on realistic assessment of capabilities and limitations.

Key elements:

• Understanding AI strengths and weaknesses
• Matching task characteristics to AI capabilities
• Recognizing when AI is inappropriate
• Decomposing complex tasks for hybrid human-AI execution

Good delegation:

• Tasks within AI's demonstrated capabilities
• Clear success criteria exist
• Output can be verified
• Risk of error is acceptable

Poor delegation:

• Tasks requiring real-time information
• Decisions requiring accountability
• Tasks you can't verify
• High-stakes irreversible actions

Related layers: Layer 0 (Cognitive Readiness), Layer 1 (System Literacy)

Dimension 2: Description

Question: How do I specify what I want?

Core competency: Crafting clear, complete instructions that produce reliable, high-quality outputs.

Key elements:

• Role definition (who AI should act as)
• Scope bounding (what's in/out of bounds)
• Format specification (output structure)
• Decision rules (how to handle judgment calls)
• Abstraction level (detail and expertise level)

Good description:

As a senior technical writer, review this API documentation for:
1. Accuracy of code examples (test each one)
2. Completeness of parameter descriptions
3. Clarity for developers new to this API

Format: For each issue found, provide:
- Location (section/line)
- Issue type
- Current text
- Suggested revision
- Priority (High/Medium/Low)

If you're unsure whether something is an issue, include it with a "Possible" tag.

Poor description:

Review this documentation and let me know what you think.

Related layers: Layer 2 (Problem Framing), Layer 3 (Instruction Design)

Dimension 3: Discernment

Question: How do I evaluate AI output?

Core competency: Critically assessing AI outputs for accuracy, completeness, and fitness for purpose.

Key elements:

• Verification against sources
• Logic and reasoning checks
• Completeness assessment
• Bias and error detection
• Confidence calibration

Discernment practices:

VERIFICATION PROTOCOL

1. LOGIC CHECK
   □ Do conclusions follow from premises?
   □ Are there reasoning gaps?
   □ Is the argument circular?

2. FACT CHECK
   □ Verify 3+ specific claims against sources
   □ Check citations actually exist
   □ Validate quantitative claims

3. COMPLETENESS CHECK
   □ Are all requested elements present?
   □ What's notably absent?
   □ Ask AI: "What did you NOT include?"

4. CONFIDENCE ASSESSMENT
   □ What's the confidence level?
   □ Where is AI most/least certain?
   □ What would change the assessment?

Related layers: Layer 4 (Reasoning Scaffolds), Layer 5 (Evaluation & Verification)

Dimension 4: Diligence

Question: How do I systematically improve?

Core competency: Iterating on AI interactions and building improving systems over time.

Key elements:

• Systematic iteration on outputs
• Capturing learnings
• Building reusable patterns
• Workflow integration
• Continuous improvement

Diligence practices:

ITERATION PROTOCOL

1. ASSESS OUTPUT
   - What's working?
   - What needs improvement?
   - What's the specific gap?

2. DIAGNOSE CAUSE
   - Is it a delegation issue?
   - Is it a description issue?
   - Is it an AI limitation?

3. REFINE APPROACH
   - What specific change will address the gap?
   - Test one change at a time
   - Document what you learn

4. CAPTURE PATTERN
   - If this worked, document why
   - Create reusable template
   - Share with others

Related layers: Layer 6 (Workflow Integration), Layer 7 (System Governance), Layer 8 (Strategic Fluency)

4D Assessment Framework

Self-Assessment

Rate each dimension (1-5):

4D SELF-ASSESSMENT

DELEGATION (Task Selection)
1 - Struggle to identify appropriate AI tasks
2 - Sometimes pick tasks AI handles poorly
3 - Generally good task selection
4 - Consistently good task-capability matching
5 - Expert at decomposing complex tasks for AI

Score: ___

DESCRIPTION (Instructions)
1 - Prompts are vague, results inconsistent
2 - Basic structure but missing elements
3 - Good prompts with role, scope, format
4 - Consistently well-structured prompts
5 - Prompts are reusable specifications

Score: ___

DISCERNMENT (Verification)
1 - Accept AI output without verification
2 - Occasional spot checks
3 - Regular verification of key claims
4 - Systematic verification protocol
5 - Comprehensive multi-gate verification

Score: ___

DILIGENCE (Improvement)
1 - Same approach regardless of results
2 - Occasional iteration when problems obvious
3 - Regular iteration and improvement
4 - Systematic capture of learnings
5 - Documented workflows with metrics

Score: ___

TOTAL: ___ / 20

Interpretation:
4-8:   Beginner - Focus on fundamentals
9-12:  Developing - Build systematic practices
13-16: Proficient - Refine and specialize
17-20: Expert - Share and scale

Gap Analysis

When AI isn't working well:

4D GAP ANALYSIS

Symptom: [What's going wrong]

DELEGATION CHECK:
□ Was this an appropriate task for AI?
□ Should it have been decomposed differently?
□ Did I overestimate AI capability?
Gap found: [Yes/No] Details: ___

DESCRIPTION CHECK:
□ Were instructions clear and complete?
□ Was the format specified?
□ Were decision rules explicit?
Gap found: [Yes/No] Details: ___

DISCERNMENT CHECK:
□ Did I verify appropriately?
□ What did I miss?
□ Was my confidence calibrated?
Gap found: [Yes/No] Details: ___

DILIGENCE CHECK:
□ Did I iterate effectively?
□ Did I capture learnings?
□ Is there a pattern to improve?
Gap found: [Yes/No] Details: ___

Primary gap: _______________
Remediation: _______________

Dimension Interactions

How Dimensions Compound

Strong Delegation + Weak Description = Right task, wrong execution
Strong Description + Weak Discernment = Good output, unverified errors
Strong Discernment + Weak Diligence = Catches errors, doesn't improve
Strong Diligence + Weak Delegation = Improving at wrong tasks

Development Sequence

Recommended progression:

1. Start with Discernment - Learn to evaluate output before trusting it
2. Build Description - Learn to get better output to evaluate
3. Develop Delegation - Learn what AI can/cannot do well
4. Add Diligence - Build systems that improve over time

Practices

4D Daily Check

TODAY'S AI INTERACTIONS

Task 1: [Description]
- Delegation: Was this appropriate? [Y/N]
- Description: Were instructions clear? [Y/N]
- Discernment: Did I verify adequately? [Y/N]
- Diligence: What did I learn? [Notes]

Task 2: [Description]
...

Pattern to improve: [What I'll do differently]

4D Prompt Review

Before running important prompts:

4D PROMPT CHECK

□ DELEGATION: Is this task appropriate for AI?
□ DESCRIPTION: Are instructions complete (role, scope, format, rules)?
□ DISCERNMENT: How will I verify the output?
□ DILIGENCE: How will I capture what I learn?

Assessment Criteria

4D Framework Mastery When:

• Can assess own AI use across all four dimensions
• Diagnoses problems by identifying which dimension is weak
• Has systematic practices for each dimension
• Dimensions work together fluidly
• Can teach framework to others

Related Skills

Each dimension maps to AI Fluency layers:

Learn More

• 4D Assessment Templates
• Dimension Development Exercises