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:
markdown
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:
markdown
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:
markdown
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:
markdown
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):
markdown
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:
markdown
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
code
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:
- •Start with Discernment - Learn to evaluate output before trusting it
- •Build Description - Learn to get better output to evaluate
- •Develop Delegation - Learn what AI can/cannot do well
- •Add Diligence - Build systems that improve over time
Practices
4D Daily Check
markdown
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:
markdown
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:
| Dimension | Primary Layers | Key Skills |
|---|---|---|
| Delegation | 0, 1 | ai-cognitive-readiness, ai-system-literacy |
| Description | 2, 3 | ai-problem-framing, ai-instruction-design |
| Discernment | 4, 5 | ai-reasoning-scaffolds, ai-evaluation-verification |
| Diligence | 6, 7, 8 | ai-workflow-integration, ai-system-governance, ai-strategic-fluency |