AI Collaborate Teaching

Purpose

Enable educators to design co-learning experiences where AI is a bidirectional learning partner following the Three Roles Framework, not just autocomplete. This skill helps:

• Teach "Specs Are the New Syntax" as the PRIMARY skill (not code-writing)
• Design lessons that emphasize specification-first, co-learning with AI, and validation-before-trust
• Establish patterns for AI pair programming in education
• Build AI tool literacy (capabilities, limitations, verification), with explicit spec → generate → validate loops
• Demonstrate the Three Roles Framework (AI as Teacher/Student/Co-Worker)
• Show bidirectional learning (human teaches AI, AI teaches human)
• Create ethical guidelines for responsible AI use
• Assess appropriate balance of AI integration in curriculum

The Three Roles Framework (Section IIa Stage 2, Constitution v5.0.0)

CRITICAL: All co-learning content MUST demonstrate this framework (per Section IIa Stage 2 forcing functions):

AI's Three Roles:

1. Teacher: Suggests patterns, architectures, best practices students may not know
2. Student: Learns from student's domain expertise, feedback, corrections
3. Co-Worker: Collaborates as peer, not subordinate

Human's Three Roles:

1. Teacher: Guides AI through clear specifications, provides domain knowledge
2. Student: Learns from AI's suggestions, explores new patterns
3. Orchestrator: Designs collaboration strategy, makes final decisions

The Convergence Loop

Required Pattern for All AI-Integrated Lessons:

┌─────────────────────────────────────────────────────────┐
│  1. Human specifies intent (with context/constraints)   │
│  2. AI suggests approach (may include new patterns)     │
│  3. Human evaluates AND LEARNS ("I hadn't thought of X")│
│  4. AI learns from feedback (adapts to preferences)     │
│  5. CONVERGE on optimal solution (better than either    │
│     could produce alone)                                 │
└─────────────────────────────────────────────────────────┘

Content Requirements:

• ✅ At least ONE instance per lesson where student learns FROM AI's suggestion
• ✅ At least ONE instance where AI adapts TO student's feedback
• ✅ Convergence through iteration (not "perfect on first try")
• ✅ Both parties contributing unique value
• ❌ NEVER present AI as passive tool awaiting commands
• ❌ NEVER show only human teaching AI (one-way instruction)
• ❌ NEVER hide what student learns from AI's approaches

Relationship to Graduated Teaching Pattern (Constitution Principle 13)

This skill complements the graduated teaching pattern:

Graduated Teaching Pattern (Constitution Principle 13) defines WHAT book teaches vs WHAT AI handles:

• Tier 1: Book teaches foundational concepts (stable, won't change)
• Tier 2: AI companion handles complex execution (student specifies, AI executes)
• Tier 3: AI orchestration at scale (10+ items, multi-step workflows)

This skill (AI Collaborate Learning) defines HOW students use AI during learning:

• When AI is involved (from Pattern Tier 2+), students use AI collaboration patterns (explainer, debugger, pair programmer)
• Balance AI-assisted work with independent verification (40/40/20 model)
• Apply ethical guidelines and verification strategies

In Practice:

1. Book teaches Markdown # headings (Tier 1 - foundational)
   → Students practice manually
   → No AI collaboration patterns needed yet

2. Students learn Markdown tables (Tier 2 - complex syntax)
   → AI companion handles table generation
   → Now apply AI collaboration patterns from this skill:
     - Student specifies table requirements
     - AI generates table
     - Student validates output
     - Student can ask AI to explain syntax (AI as Explainer)

3. Students convert 10 documents (Tier 3 - orchestration)
   → AI orchestrates batch conversion
   → Apply AI pair programming pattern (AI as Pair Programmer)
   → Maintain 40/40/20 balance with verification checkpoints

Key Integration Points:

With 4-Layer Method (Section IIa):

• Layer 1 (Manual practice): Minimal AI collaboration — build independent capability first
• Layer 2-4 (AI-assisted onward): Apply this skill's collaboration patterns

With Graduated Teaching (Principle 2):

• Tier 1 (Foundational): Book teaches directly — minimal AI patterns needed
• Tier 2 (Complex): AI companion handles — apply this skill's collaboration patterns
• Tier 3 (Scale): AI orchestration — full pair programming with strategic oversight

Refer to Section IIa (4-Layer Method) and Principle 2 (Graduated Teaching) for decisions about WHEN and WHAT. Use this skill for HOW students collaborate with AI effectively.

When to Activate

Use this skill when:

• Designing programming courses that integrate AI coding assistants
• Teaching students to use AI tools (ChatGPT, GitHub Copilot, Claude) effectively
• Creating prompt engineering curriculum or exercises
• Establishing policies for AI use in programming education
• Balancing AI assistance with independent skill development
• Assessing whether AI integration enhances or hinders learning
• Educators ask about "AI in teaching", "prompt engineering pedagogy", "AI pair programming", "AI tool literacy"
• Reviewing existing AI-integrated curricula for improvements

Process

Step 1: Understand the Educational Context

When a request comes in to integrate AI into programming education, first clarify:

• What programming topic or course? (Intro to Python, web development, data structures, etc.)
• What is the student level? (Complete beginners, intermediate, advanced)
• What AI tools are available? (ChatGPT, GitHub Copilot, Claude, other)
• What are the learning objectives? (What should students be able to do?)
• What foundational skills must be built independently? (Core concepts that shouldn't use AI)
• What ethical concerns exist? (Academic integrity, over-reliance, attribution)

Step 2: Review Prompt Engineering Pedagogy

Learn how to teach students to craft effective prompts: 📖 reference/prompt-engineering-pedagogy.md

This document covers:

• Four Prompt Competencies: Context setting, constraint specification, output format, iterative refinement
• Teaching Prompt Quality: Clarity, specificity, context completeness, testability
• Scaffolding Strategies: Templates (beginner), critique (intermediate), independent crafting (advanced)
• Common Anti-Patterns: Vague requests, assuming AI knows context, overloading prompts, passive acceptance
• Assessment Strategies: Prompt journals, prompt challenges, peer review

Key Insight: Prompt engineering is about effective communication, problem specification, and critical evaluation - all valuable software engineering skills.

Step 3: Design AI Pair Programming Patterns

Review how students can work with AI as a collaborative partner: 📖 reference/ai-pair-programming-patterns.md

This document covers five patterns:

• Pattern 1: AI as Explainer - Student inquires, AI clarifies concepts
• Pattern 2: AI as Debugger - Student reports bugs, AI helps diagnose
• Pattern 3: AI as Code Reviewer - Student writes code, AI provides feedback
• Pattern 4: AI as Pair Programmer - Student and AI co-create code incrementally
• Pattern 5: AI as Hypothesis Validator - Student forms hypotheses, AI confirms/refutes

Critical Balance: Student should understand and own all code, not just copy-paste AI output.

Teaching Strategies:

• Scaffold from guided templates to independent use
• Require students to explain all code (even AI-generated)
• Include AI-free checkpoints to verify learning
• Balance assistance with independent struggle

Step 4: Build AI Tool Literacy

Teach students to understand AI capabilities and limitations: 📖 reference/ai-tool-literacy.md

This document covers:

• What AI Does Well: Pattern recognition, code generation, explanation, refactoring, debugging common issues
• What AI Does Poorly: Complex domain logic, system design, originality, understanding unstated context, comprehensive security
• Conceptual Understanding: AI is pattern recognition from training data, not logical reasoning
• Verification Strategies: Read/understand, test thoroughly, code review, cross-check documentation, run and observe
• When to Trust: High confidence for well-known patterns, low confidence for security/performance/complex logic
• Recognizing Biases: Recency, popularity, correctness, cultural, representation biases

Key Principle: Trust, but verify - always.

Step 5: Establish Ethical Guidelines

Create clear ethical frameworks for AI use: 📖 reference/ethical-ai-use.md

This document covers seven ethical principles:

1. Honesty and Transparency: Disclose AI assistance
2. Academic Integrity: AI enhances learning, doesn't substitute for it
3. Attribution and Credit: Give credit where due
4. Understanding Over Outputs: Never submit code you don't understand
5. Bias Awareness: Recognize AI limitations and biases
6. Over-Reliance Prevention: Maintain independent coding ability
7. Professional Responsibility: You're accountable for all code

Teaching Strategies:

• Set explicit policies early (Week 1)
• Discuss ethical dilemmas regularly
• Model ethical AI use
• Require process documentation (when/why AI was used)
• Include AI-free assessments periodically

Step 6: Design AI-Integrated Lesson

Use the lesson template to structure AI integration: 📄 templates/ai-lesson-template.yml

The template includes:

• Lesson Metadata: Topic, duration, audience, AI integration level
• Learning Objectives: With AI role specified for each
• Foundational vs. AI-Assisted Skills: What must be learned independently vs. with AI help
• Lesson Phases:
  • Introduction (no AI): Motivation and prerequisites
  • Foundation (no AI): Build core concepts independently first
  • AI-Assisted Exploration (with AI): Practice and explore with scaffolding
  • Independent Consolidation (no AI): Verify learning without AI
  • Wrap-Up: Reflection and discussion
• AI Integration Strategy: Tools, guidelines, prompt templates, disclosure requirements
• Balance Assessment: 40% foundational / 40% AI-assisted / 20% verification (target ratio)
• Ethical Considerations: Policies, prohibited actions, verification requirements

Key Structure: Always start with independent foundation, allow AI assistance with scaffolding, verify learning independently.

Step 7: Create Effective Prompt Templates

Provide students with templates for different tasks: 📄 templates/prompt-design-template.md

This template provides structures for:

• Basic Prompt Structure: Context + Task + Constraints
• Detailed Prompt Template: With focus areas and output format specs
• Task-Specific Templates: Code generation, explanation, debugging, code review, alternatives
• Anti-Patterns: What to avoid
• Prompt Quality Checklist: Verify before submission

Teaching Approach: Start with templates, gradually remove scaffolding as students gain expertise.

Step 8: Assess AI Integration Balance

Once a lesson is designed, validate the AI integration:

python .claude/skills/ai-augmented-teaching/scripts/assess-ai-integration.py lesson-plan.yml

The script assesses:

• ✅ Balance: Is the ratio appropriate (foundation/AI-assisted/verification)?
• ✅ Foundational Skills: Are core skills protected from AI assistance?
• ✅ Verification: Are there checkpoints to test learning without AI?
• ✅ Ethical Guidelines: Are disclosure, understanding, and verification required?

Interpret Results:

• Overall Score: 90+ (Excellent), 75-89 (Good), 60-74 (Needs Improvement), <60 (Poor)
• Balance Issues: Adjust percentages if too much/little AI assistance
• Missing Verification: Add independent checkpoints
• Ethical Gaps: Include disclosure requirements, understanding checks

If score is low:

1. Review recommendations
2. Adjust lesson phases (add independent work or verification)
3. Clarify foundational vs. AI-assisted skills
4. Add ethical guidelines
5. Re-assess until score improves

Step 9: Validate Prompt Quality

For prompt engineering exercises, validate prompt quality:

python .claude/skills/ai-augmented-teaching/scripts/validate-prompts.py prompts.yml

The script checks:

• Clarity: Is the prompt specific and clear?
• Context: Does it provide adequate background?
• Task Specification: Is the requested task explicit?
• Testability: Can the output be verified?
• Constraints: Are requirements and limitations specified?

Interpret Results:

• Quality Score: 85+ (Excellent), 70-84 (Good), 50-69 (Needs Improvement), <50 (Poor)
• Suggestions: Specific improvements for each prompt
• Common Issues: Vague language, missing context, unclear tasks

Use for:

• Evaluating student-written prompts
• Improving prompt templates
• Teaching prompt quality criteria

Step 10: Iterate and Refine

After teaching with AI integration:

1. Gather Feedback: What worked? What didn't?
2. Assess Learning: Did students achieve objectives independently?
3. Check for Over-Reliance: Can students code without AI?
4. Review Ethical Use: Were guidelines followed?
5. Adjust Balance: Increase/decrease AI assistance based on outcomes

Output Format

Present AI-integrated lesson plans following the ai-lesson-template.yml structure:

lesson_metadata:
  title: "Lesson Title"
  topic: "Programming Topic"
  duration: "90 minutes"
  ai_integration_level: "Medium"

learning_objectives:
  - statement: "Students will be able to [action]"
    ai_role: "Explainer | Pair Programmer | Code Reviewer | None"

foundational_skills_focus:
  - "Core skill 1 (no AI)"
  - "Core skill 2 (no AI)"

ai_assisted_skills_focus:
  - "Advanced skill 1 (with AI)"
  - "Advanced skill 2 (with AI)"

phases:
  - phase_name: "Foundation (Independent)"
    ai_usage: "None"
    activities: [...]

  - phase_name: "AI-Assisted Exploration"
    ai_usage: "Encouraged"
    activities: [...]

  - phase_name: "Independent Consolidation"
    ai_usage: "None"
    activities: [...]

ai_assistance_balance:
  foundational_work_percentage: 40
  ai_assisted_work_percentage: 40
  independent_verification_percentage: 20

Acceptance Checks

• Spectrum tag specified for the lesson: Assisted | Driven | Native
• Spec → Generate → Validate loop outlined for AI usage
• At least one “verification prompt” included to force the model to explain/test its own output

Verification prompt examples

- “Explain why this output satisfies the acceptance criteria from the spec.”
- “Generate unit tests that would fail if requirement X is not met.”
- “List assumptions you made; propose a test to verify each.”

Examples

Example 1: Intro to Python Functions (Beginner)

Context: Teaching functions to absolute beginners

AI Integration Strategy:

lesson_metadata:
  title: "Introduction to Python Functions"
  duration: "90 minutes"
  target_audience: "Beginners"
  ai_integration_level: "Low"

foundational_skills_focus:
  - "Understanding function syntax (def, parameters, return)"
  - "Tracing function execution mentally"
  - "Writing simple functions independently"

ai_assisted_skills_focus:
  - "Exploring function variations"
  - "Generating test cases"
  - "Getting alternative implementations"

phases:
  - phase_name: "Foundation (30 min, No AI)"
    activities:
      - Introduce function concepts (lecture)
      - Work through examples on board
      - Students write 3 simple functions independently
      - Quick comprehension check

  - phase_name: "AI-Assisted Practice (40 min)"
    activities:
      - Students use AI to explain functions they don't understand
      - Request AI help generating test cases
      - Ask AI for alternative approaches
      - All AI usage must be documented

  - phase_name: "Independent Verification (15 min, No AI)"
    activities:
      - Write 2 functions without AI assistance
      - Explain what each function does
      - Prove they can code functions independently

ai_assistance_balance:
  foundational: 40%
  ai_assisted: 45%
  verification: 15%

Rationale: Beginners need strong foundation before AI assistance. Mostly independent work.

---

Example 2: Web API Integration (Intermediate)

Context: Teaching how to integrate external APIs

AI Integration Strategy:

lesson_metadata:
  title: "Integrating REST APIs in Python"
  duration: "2 hours"
  target_audience: "Intermediate"
  ai_integration_level: "High"

foundational_skills_focus:
  - "Understanding HTTP methods (GET, POST, PUT, DELETE)"
  - "Reading API documentation"
  - "Handling JSON responses"

ai_assisted_skills_focus:
  - "Crafting API requests with authentication"
  - "Error handling for network issues"
  - "Building robust API clients"

phases:
  - phase_name: "Foundation (25 min, No AI)"
    activities:
      - Review HTTP basics
      - Demonstrate simple API call with requests library
      - Students make first API call independently

  - phase_name: "AI-Assisted Building (60 min)"
    activities:
      - Use AI as pair programmer to build API client
      - Request AI help with authentication patterns
      - Ask AI to suggest error handling strategies
      - Students build incrementally with AI assistance

  - phase_name: "Independent Consolidation (25 min, No AI)"
    activities:
      - Extend API client with new endpoint (no AI)
      - Debug intentionally broken API call
      - Explain all code including AI-generated parts

ai_assistance_balance:
  foundational: 25%
  ai_assisted: 55%
  verification: 20%

Rationale: Intermediate students can handle more AI integration. Foundation is brief since they know Python basics.

---

Example 3: Prompt Engineering Bootcamp (Advanced)

Context: Teaching prompt engineering as a skill

AI Integration Strategy:

lesson_metadata:
  title: "Mastering Prompt Engineering for Code"
  duration: "3 hours"
  target_audience: "Advanced"
  ai_integration_level: "High"

foundational_skills_focus:
  - "Understanding prompt structure (context/task/constraints)"
  - "Identifying vague vs. specific prompts"
  - "Recognizing AI capabilities and limitations"

ai_assisted_skills_focus:
  - "Iterative prompt refinement"
  - "Crafting complex multi-step prompts"
  - "Effective code review requests"

phases:
  - phase_name: "Prompt Quality Foundation (30 min, No AI)"
    activities:
      - Analyze good vs. bad prompts
      - Practice prompt critique
      - Learn quality criteria (clarity, context, testability)

  - phase_name: "Iterative Prompt Design (90 min, With AI)"
    activities:
      - Students write prompts for complex tasks
      - Test prompts with AI, evaluate outputs
      - Refine prompts based on results
      - Compare approaches with peers

  - phase_name: "Prompt Challenge (30 min, No AI first)"
    activities:
      - Design prompts for given scenarios (no AI)
      - Then test prompts with AI
      - Evaluate: Did prompts produce useful outputs?
      - Reflect on prompt quality and effectiveness

ai_assistance_balance:
  foundational: 20%
  ai_assisted: 60%
  verification: 20%

Rationale: Advanced students learning prompt engineering should spend most time experimenting with AI. But they must demonstrate prompt design skills independently first.

---

Common Patterns

Pattern 1: 40/40/20 Balance (Standard)

40% Foundation (no AI): Build core skills independently
40% AI-Assisted: Practice and explore with AI support
20% Verification (no AI): Prove independent capability

Use for: Most programming lessons for intermediate students

---

Pattern 2: 60/20/20 Balance (Beginner-Heavy)

60% Foundation (no AI): Extensive independent skill-building
20% AI-Assisted: Limited, scaffolded AI use
20% Verification (no AI): Ensure basics are solid

Use for: Absolute beginners, core foundational concepts

---

Pattern 3: 25/55/20 Balance (Advanced Integration)

25% Foundation (no AI): Brief independent practice
55% AI-Assisted: Heavy AI collaboration
20% Verification (no AI): Confirm understanding

Use for: Advanced students, exploring new libraries/frameworks

---

Troubleshooting

Assessment Shows Poor Balance (<60 score)

Problem: assess-ai-integration.py reports low score

Common Issues:

1. Too much AI assistance (>60%) - Students won't build independent skills
2. Too little verification (<15%) - No way to confirm learning
3. No foundational phase - Students use AI from the start
4. Missing ethical guidelines

Solutions:

1. Add foundational phase (no AI) at the beginning
2. Reduce AI-assisted percentage to 30-50%
3. Add independent verification phase at end
4. Include disclosure requirements and ethical guidelines
5. Re-assess until score improves to 75+

---

Students Over-Rely on AI

Problem: Students can't code without AI assistance

Indicators:

• Panic when AI unavailable
• Can't explain AI-generated code
• Performance drops significantly on AI-free assessments

Solutions:

1. Increase AI-Free Time: More foundational and verification phases
2. 20-Minute Rule: Students must try independently for 20 min before AI
3. Progressive Independence: Gradually reduce AI assistance over semester
4. Regular AI-Free Assessments: Verify retention of skills

---

Prompts Are Low Quality (<50 score)

Problem: validate-prompts.py reports poor quality prompts

Common Issues:

• Too vague: "Write code for sorting"
• No context: "Fix this" [paste code]
• No testability: Can't verify if output is correct
• Missing constraints: No requirements specified

Solutions:

1. Teach Prompt Structure: Context + Task + Constraints + Output Format
2. Provide Templates: Scaffold with fill-in-the-blank templates
3. Prompt Critique Practice: Analyze good vs. bad prompts
4. Iterative Refinement: Show how to improve prompts based on results

---

Ethical Violations Occur

Problem: Students use AI without disclosure, submit code they don't understand

Prevention:

1. Set Policy Early: Week 1, explicit guidelines
2. Require Documentation: Students log all AI use
3. Explanation Requirement: Must explain all code (including AI-generated)
4. AI-Free Assessments: Periodically verify independent capability
5. Consequences: Clear penalties for violations

---

Teaching Agentic AI and Advanced Topics

As curriculum evolves to include agentic AI systems and Model Context Protocol (MCP), teaching strategies shift:

Special Considerations for Agentic AI

Agentic AI differs from traditional AI assistance:

• Students are designing AGENTS (goal-seeking systems), not just using AI as a code generator
• Agency and autonomy introduce new concepts: agent goals, decision-making, state management, tool selection
• Students must understand agent behavior at a deeper level (not just "give it a prompt")

Teaching Agentic AI Effectively:

1. Start with Agent Concepts (Not Just Prompting)

   • Begin with what agents ARE and why they differ from traditional AI use
   • Use diagrams showing agent loops: perceive → decide → act → repeat
   • Compare agents with traditional chatbots (students often conflate them)

2. Build Agent Design Gradually

   • First agents: simple goal-seeking with 2-3 available tools
   • Mid-level: agents with state management and complex goals
   • Advanced: agent orchestration and multi-agent systems

3. Include Failure Analysis

   • Agents often fail or loop - teach students to recognize and debug these
   • Log analysis exercises: "Why did the agent pick the wrong tool?"
   • Improvement exercises: "How would you change the goal/tools to fix this?"

4. Emphasize Agent Testing and Safety

   • Simple prompts can work fine; complex agents need careful testing
   • Teach students to set boundaries and constraints for agents
   • Include cost monitoring (API calls can add up with agents!)

5. Real-World Agent Projects

   • Research assistant agent
   • Data processing agent
   • System administration agent
   • Customer support agent
   • Each demonstrates different agent patterns and challenges

Special Considerations for MCP (Model Context Protocol)

MCP extends traditional AI assistance:

• MCP servers provide tools/resources that models can access
• Students learn to integrate external capabilities into AI systems
• Bridge between application development and AI enhancement

Teaching MCP Effectively:

1. Start with Architecture Understanding

   • Draw diagrams: Client ← Protocol → Server
   • Explain what servers can provide (tools, resources, data access)
   • Compare with traditional APIs (similar but bidirectional communication)

2. Learn Existing MCP Servers First

   • Install and integrate established MCP servers
   • Understand how applications use MCP
   • Build confidence with known tools before creating custom ones

3. Build Custom MCP Servers

   • Start simple: single-purpose server with 2-3 tools
   • Progress to complex: multi-tool servers with state management
   • Industry example: build an MCP server for your domain (database access, API wrapper, etc.)

4. Integrate MCP + Agents

   • Advanced students can build agents that use MCP servers
   • Students appreciate how MCP provides reliable tool access for agents
   • Real problem-solving: agent + MCP creates powerful combinations

5. Emphasize Reusability

   • Well-designed MCP servers are reusable across projects
   • Teach documentation: others should be able to use your server
   • Portfolio value: publishing MCP servers shows engineering maturity

---

Integration with Other Skills

This skill works well with:

→ learning-objectives skill: Define what students should achieve, then decide what AI role supports those objectives

→ exercise-designer skill: Create exercises that balance AI assistance with independent practice

→ assessment-builder skill: Design assessments measuring understanding (not just code completion)

→ code-example-generator skill: Generate examples, then teach students to use AI similarly

---

Tips for Success

1. Start with Foundation: Always build core skills independently before AI
2. Balance is Critical: 40/40/20 is a good starting ratio
3. Verify Learning: AI-free checkpoints are non-negotiable
4. Teach Verification: Students must test and understand AI outputs
5. Model Ethical Use: Demonstrate how YOU use AI responsibly
6. Iterate Prompts: First prompts are rarely perfect
7. Document Everything: Require students to log AI usage
8. Maintain Independence: Periodic AI-free work ensures skills remain
9. Discuss Ethics Often: Not just Week 1 - ongoing conversations
10. Adapt to Context: Beginners need more foundation, advanced students can handle more AI

---

Ready to design AI-integrated curriculum? Provide:

• Programming topic and level
• Student audience (beginner/intermediate/advanced)
• Available AI tools
• Learning objectives
• Current concerns (over-reliance, academic integrity, etc.)

Or share an existing lesson plan and I'll assess AI integration balance and suggest improvements!