Think Tank

Quick Start

If user intent is vague: Use l'entonnoir questions → Validate incrementally → Apply YAGNI

If analysis is needed: Route to reasoning framework → Execute technique → Deliver insight

If unsure what to use: Ask "What are you trying to do?" with options

Navigation

PATTERN: L'Entonnoir Workflow

Collaborative exploration for vague intent:

1. Clarify Intent → Present 2-4 recognition-based options, each with tradeoffs
2. Explore Context → Quick project state check (existing patterns, stack)
3. Validate Incrementally → Present design in 200-300 word sections
4. Apply YAGNI → Strip speculative features

Question Flow

Start broad, narrow progressively:

"What type of project?"
→ "What specific capability?"
→ "What integration needs?"
→ "What constraints matter?"

Incremental Validation

Present sections one at a time:

**Architecture**

[200-300 words]

"Does this direction work, or adjust?"

[Wait for confirmation before proceeding]

YAGNI Check

For each feature proposed:

• "Is this needed now, or speculative?"
• "What's the minimum to achieve the goal?"
• "What can we defer?"

PATTERN: Intent Router

Route to reasoning framework when analysis is the goal:

When Multiple Apply

Select the most appropriate technique based on analysis of the user's intent. Consider:

1. What is the core problem or question?
2. What outcome does the user need?
3. Which technique directly addresses this?

Trust your judgment—the AI agent can infer and select appropriately.

PATTERN: Technique Reference

5 Whys

Drill to root cause.

Output: Problem → 5 levels of "why" → Root cause → Intervention

Pareto

Find vital few (~20% driving ~80%).

Output: High-impact factors to focus on → Low-impact to deprioritize

Inversion

What would guarantee failure?

Output: Failure modes → Avoidance strategies → "Never do" boundaries

SWOT

Internal strengths/weaknesses, external opportunities/threats.

Output: 4-quadrant map → SO/WO/ST/WT strategic moves

10/10/10

Evaluate across time horizons.

Output: Each option at 10min / 10months / 10years → Time conflicts → Recommendation

The One Thing

Single highest-leverage action.

Output: Goal → Candidate actions → Leverage point → Immediate next action

Occam's Razor

Simplest explanation fitting all facts.

Output: Explanations with assumptions → Evidence check → Simplest valid

Second Order

Consequences of consequences.

Output: First-order → Second-order → Third-order → Revised assessment

Eisenhower Matrix

Urgent/important prioritization.

Output: Q1 (Do First) → Q2 (Schedule) → Q3 (Delegate) → Q4 (Eliminate)

Opportunity Cost

What you give up.

Output: Resources required → Best alternative uses → True cost → Verdict

Via Negativa

Improve by removing.

Output: Current state → Subtraction candidates → What to keep → After subtraction

First Principles

Rebuild from fundamentals.

Output: Assumptions challenged → Fundamental truths → Rebuilt understanding → New possibilities

ANTI-PATTERN: Common Mistakes

Mistake 1: Questions Without Tradeoffs

❌ "What framework do you want?"

✅ "What type of project? (a) API, (b) Web app, (c) CLI - each has different stack implications"

Mistake 2: Validating Too Much at Once

❌ "Here's the complete design"

✅ Present 200-300 words at a time, confirm each section

Mistake 3: Accepting Speculative Features

❌ "We might need X later"

✅ "Build only what's needed now. Later = later."

Mistake 4: Analysis Without Action

❌ "Here's a SWOT analysis"

✅ SWOT → Strategic moves → What to do next

Mistake 5: Skipping Context

❌ "What database?" without checking existing stack

✅ Check existing patterns first, then propose options

Recognition Questions

Delta Standard: Instructions assume basic reasoning competence. Don't explain what reasoning entails—apply it.