
ai-reasoning-scaffolds
Use AI as a thinking partner by injecting checklists, reasoning trees, critique loops, and stepwise
Overview
AI Reasoning Scaffolds is Layer 4 of AI fluency—the ability to structure AI's reasoning process rather than just its output. This transforms AI from a generator into a thinking partner.
Core Principle: AI follows the reasoning structure you provide. No scaffold = no reliable reasoning.
Fluency Signal: AI outputs show internal coherence and traceable logic.
When to Use This Skill
- Complex analysis requiring multiple considerations
- Decision-making with trade-offs
- Problems requiring systematic evaluation
- When AI output lacks depth or misses important factors
- When you need to understand AI's reasoning path
Scaffold Types
1. Checklists
Purpose: Ensure systematic coverage of required elements.
Pattern:
Before providing your analysis, work through this checklist:
□ Identify the main claim
□ List supporting evidence
□ Identify gaps in evidence
□ Consider counter-arguments
□ Assess source credibility
□ Note any unstated assumptions
□ Rate confidence level (High/Medium/Low)
Then synthesize your findings into a response.
When to use:
- Quality assurance tasks
- Review processes
- Comprehensive analysis
- Compliance checking
2. Reasoning Trees
Purpose: Guide systematic exploration of options.
Pattern:
Analyze this problem using the following decision tree:
1. First, classify the problem type:
- If performance issue → Go to Branch A
- If feature request → Go to Branch B
- If unclear → Ask clarifying question
Branch A (Performance):
2. Identify the bottleneck layer:
- Database → Check query optimization
- Application → Check algorithm complexity
- Network → Check payload size
3. For each identified issue, propose solutions ranked by effort/impact
Branch B (Feature):
2. Assess alignment with roadmap...
[continue structure]
When to use:
- Diagnostic problems
- Classification tasks
- Decision processes
- Troubleshooting
3. Stepwise Reasoning
Purpose: Force explicit thinking through steps.
Pattern:
Solve this problem step by step. Show your work for each step:
Step 1: State the problem in your own words
[Your response]
Step 2: Identify what information is given
[Your response]
Step 3: Identify what information is needed
[Your response]
Step 4: Determine the approach
[Your response]
Step 5: Execute the approach
[Your response]
Step 6: Verify the result
[Your response]
Step 7: State the conclusion
[Your response]
When to use:
- Mathematical or logical problems
- Complex analysis
- Debugging
- Audit trails needed
4. Critique Loops
Purpose: Generate opposing perspectives.
Pattern:
Analyze this proposal using a structured critique:
ROUND 1 - ADVOCATE:
Present the strongest case FOR this proposal. What are the benefits?
ROUND 2 - CRITIC:
Now argue AGAINST the proposal. What are the risks and weaknesses?
ROUND 3 - SYNTHESIS:
Reconcile these perspectives. What's the balanced view? What conditions would make this proposal succeed or fail?
When to use:
- Evaluating options
- Risk assessment
- Stress-testing ideas
- Avoiding confirmation bias
5. Multi-Perspective Analysis
Purpose: Examine from multiple stakeholder viewpoints.
Pattern:
Analyze this decision from multiple perspectives:
PERSPECTIVE 1 - Customer:
- What do they gain?
- What concerns would they have?
PERSPECTIVE 2 - Engineering:
- What's the implementation complexity?
- What are the technical risks?
PERSPECTIVE 3 - Business:
- What's the revenue impact?
- What's the competitive implication?
PERSPECTIVE 4 - Operations:
- How does this affect support load?
- What's the maintenance burden?
SYNTHESIS:
Which perspective carries the most weight for this decision? What trade-offs are acceptable?
When to use:
- Strategic decisions
- Product decisions
- Organizational changes
- Trade-off analysis
Scaffold Design Principles
Principle 1: Explicit > Implicit
Don't assume AI will consider something—make it explicit:
Weak:
"Consider all relevant factors"
Strong:
"Consider these factors: cost, timeline, risk, team capacity, dependencies"
Principle 2: Ordered > Unordered
Sequence matters for reasoning quality:
Weak:
"Think about benefits, risks, costs, and feasibility"
Strong:
"First identify benefits, then list risks for each benefit, then estimate costs, then assess overall feasibility"
Principle 3: Constrained > Open
Bounded options produce better reasoning:
Weak:
"Rate how confident you are"
Strong:
"Rate confidence as: High (>80% sure), Medium (50-80%), Low (<50%). State the main uncertainty."
Principle 4: Observable > Hidden
Request visible reasoning:
Weak:
"Give me your best answer"
Strong:
"Show your reasoning process, then give your conclusion. I need to understand how you arrived at it."
Delegation Patterns
Delegate Sub-Reasoning
For complex problems, delegate specific reasoning tasks:
Main problem: [Complex question]
I'll work through this systematically. For each sub-question, provide your reasoning:
Sub-question 1: [Bounded aspect of the problem]
[AI responds with focused analysis]
Sub-question 2: [Another aspect]
[AI responds]
Now I'll synthesize these inputs into my decision.
Chain Reasoning Across Turns
Turn 1: "Analyze the current state of X"
[AI provides analysis]
Turn 2: "Given your analysis, what are the top 3 strategic options?"
[AI generates options building on prior analysis]
Turn 3: "For option 2, do a deeper risk analysis"
[AI drills into specific option]
Turn 4: "Synthesize our conversation into a recommendation"
[AI produces coherent output from the reasoning chain]
Practices
Chain-of-Reasoning Prompting
Add reasoning scaffolds to any prompt:
[Your original prompt]
Think through this step by step:
1. First, understand what's being asked
2. Identify the key factors that matter
3. Consider the implications of each factor
4. Weigh trade-offs if any exist
5. Form your conclusion
6. Verify your reasoning makes sense
Show your work for steps 1-5, then provide your final answer.
Counter-Argument Generation
Build in skepticism:
[Your original prompt]
After your initial response, challenge it:
- What's the strongest argument against your conclusion?
- What assumption, if wrong, would invalidate your reasoning?
- What information would make you change your answer?
Self-Critique Prompting
[Your original prompt]
After your response:
SELF-CRITIQUE:
- What did you do well?
- What might be wrong or incomplete?
- What would you do differently with more time?
- Confidence level: High/Medium/Low and why
Assessment Criteria
Layer 4 Complete When:
- Regularly uses structured reasoning scaffolds
- AI outputs show explicit reasoning steps
- Can delegate sub-reasoning without losing coherence
- Uses critique loops to stress-test AI conclusions
- Has created reusable reasoning templates
Common Scaffold Failures
Failure 1: No Scaffold for Complex Tasks
Wrong: "Analyze this business problem" Right: "Analyze this problem using: 1) Problem definition 2) Root cause analysis 3) Option generation 4) Option evaluation 5) Recommendation"
Failure 2: Scaffold Too Loose
Wrong: "Think about pros and cons" Right: "List exactly 3 pros and 3 cons, ranked by importance, with one sentence explanation each"
Failure 3: Missing Verification Step
Wrong: "Solve this calculation" Right: "Solve this calculation, then verify by working backward from your answer"
Failure 4: Reasoning Without Output
Wrong: "Think through this carefully" Right: "Think through this carefully and show your reasoning at each step"
Related Skills
- ai-instruction-design — Format for scaffolds
- ai-evaluation-verification — Verifying scaffold outputs
- ai-problem-framing — Decomposition for delegation