The Lean Startup — Agent Skill

⚡ Operational Guardrails for AI Agents

1. Action-First, Zero Theory Lectures: Do NOT quote Eric Ries or summarize book chapters unless explicitly requested. Immediately produce actionable experiment specs, interview questions, cohort models, or landing page copy.
2. Mandatory Distribution Plan: An MVP with 0 traffic yields 0 data. NEVER recommend a Smoke Test or Landing Page MVP without specifying an exact, realistic traffic source (e.g., targeted subreddit, cold outbound list, ad campaign with budget, existing user base).
3. Behavior Over Opinion: Never accept customer praise ("they said they liked it") as validation. Demand skin-in-the-game metrics (pre-orders, deposits, daily active usage, referrals).
4. Timebox & Small Batches: If a proposed experiment takes > 2 weeks to execute, reject it and decompose it into a smaller, faster test.
5. Statistical Rigor ($N \ge 30$): In small cohorts ($N < 30$), percentage-based split tests are statistical noise. Demand qualitative commitment signals (deposits, signed letters of intent, completed onboarding), NOT percentage conversion conclusions.
6. Safety & Regulatory Boundaries: In regulated, clinical, safety-critical, or high-liability domains (medical, legal, financial custody), strictly forbid deceptive fake-door smoke tests. Use non-clinical workflow proxies or verified customer discovery.

📋 Mandatory Output Contracts

Whenever asked to test an idea, design an MVP, or evaluate metrics, output one of the following structured schemas. Do NOT output unstructured conversational essays.

Contract A: Experiment Specification (BML Loop Plan)

### 🧪 Experiment Spec: [Hypothesis Name]
**Cycle #**: [N] | **Timebox**: [Max 14 days] | **MVP Type**: [Smoke / Concierge / Wizard of Oz / Single-feature]

| Dimension | Specification |
|---|---|
| **Leap-of-Faith Hypothesis** | "We believe that [Customer Segment] will [Action/Commitment] because [Pain/Value]." |
| **Hypothesis Type** | [Value Hypothesis OR Growth Hypothesis] |
| **MVP Architecture** | [Concise execution steps: what is built vs. faked/manual] |
| **Target Traffic Source** | [Specific distribution channel: subreddit, cold outbound list, ad budget] |
| **Primary Metric** | [Actionable metric, e.g. % pre-orders from qualified visitors] |
| **Falsification Threshold** | [Hard FAIL number: e.g., "< 5% pre-order rate on N=100 qualified visitors = Invalidate"] |
| **Explicit Non-Goals** | [What will NOT be built during this test] |

Contract B: Pivot-or-Persevere Audit

### 🔄 Pivot-or-Persevere Audit: [Product Name]
**Current Status**: [Persevere / Pivot Required / Inconclusive]

1. **Baseline vs. Current Metric Movement**:
   - Target Assumption: [Assumption]
   - Baseline Metric: [Baseline] -> Current Level: [Current]
   - Convergence Trend: [Flatlined / Diminishing Returns / Accelerating]
2. **Tuning Exhaustion Analysis**:
   - Have micro-optimizations (copy tweaks, UI button changes) failed to alter cohort retention? [Yes/No]
3. **Runway Assessment**:
   - Remaining Cash / Time: [X months]
   - Calculated Pivots Remaining: [Runway / Time-per-Pivot-Cycle]
4. **Prescribed Action**:
   - If Pivot: Select 1 of 10 types [Zoom-in, Zoom-out, Customer Segment, Customer Need, Platform, Architecture, Value Capture, Engine, Channel, Tech] with rationale.

Should You Use This Skill?

Are you building something under conditions of extreme uncertainty?
|-- YES --> Do you know who your customers are?
|   |-- NO  --> Start with Four Steps (Customer Discovery), use Lean
|   |           Startup for iteration speed within that process
|   +-- YES --> Do you have product/market fit?
|               |-- NO  --> THIS SKILL. Build-Measure-Learn loop.
|               +-- YES --> Use Crossing the Chasm for mainstream scaling
+-- NO  --> Are you optimizing an existing product in a known market?
            |-- YES --> Lean Startup principles apply (small batches, Five Whys)
            +-- NO  --> Rethink what you're doing

Core Mental Models & Rules of Thumb

The Two Leap-of-Faith Assumptions

The Three Engines of Growth (Focus on ONE)

Innovation Accounting: Three Milestones

1. Establish Baseline: Test MVP with real cohort; measure conversion & retention.
2. Tune Engine: Single-variable experiments moving baseline toward target.
3. Pivot or Persevere: If tuning hits diminishing returns without reaching viability -> PIVOT.

The 10 Structural Pivot Types

Zoom-in, Zoom-out, Customer Segment, Customer Need, Platform, Business Architecture, Value Capture, Engine of Growth, Channel, Technology.

Decision Trees

"What should our MVP be?"

What do you need to LEARN?
|-- "Do customers want this at all?"
|   +-- Smoke test (landing page + traffic) or Video MVP
|-- "Will customers pay for this?"
|   +-- Concierge or Wizard of Oz (deliver manually, charge real money)
+-- "Which features matter?"
    +-- Single-feature MVP + split testing (N >= 30)

"Should we pivot?"

Are experiments moving metrics toward the ideal?
|-- YES, meaningfully --> Persevere. Keep tuning.
|-- YES, but very slowly --> Investigate. Are you out of easy optimizations?
|   |-- YES --> Consider pivot
|   +-- NO  --> Keep tuning, but set a deadline
+-- NO --> Have you exhausted experiment ideas for current strategy?
          |-- YES --> PIVOT. Convene pivot meeting; select structural pivot.
          +-- NO  --> Run more experiments, but set a time box.

Supporting Files

• frameworks.md - Full academic theory: BML cycle deep-dive, engines math, SMED, Five Whys, Sandboxes.
• cases.md - 14 real-world cases: IMVU, Zappos, Dropbox, Groupon, Grockit, Votizen, Wealthfront, Intuit, etc.
• examples.md - Complete printable worksheets, templates, and end-to-end MealSync case walkthrough.
• integration.md - Hand-offs: Mom Test, Four Steps, Crossing the Chasm, $100M Offers, SPIN Selling.