Concept of the skill

What it is: Lean Startup is Eric Ries's methodology for creating products and ventures under extreme uncertainty by running disciplined build-measure-learn loops. It treats progress as validated learning about a sustainable business, not as the amount of product shipped.

Mental model: Start with a vision, identify the leap-of-faith assumptions that must be true, choose the riskiest assumption, build the smallest ethical experiment that can test it, measure customer behavior with actionable metrics, and decide whether to pivot, persevere, stop, or run the next loop.

Why it exists: Teams often spend months building a polished product before discovering that customers do not care. Lean Startup compresses that waste by making the learning question, evidence threshold, and decision rule explicit before the build effort begins.

What it is NOT: It is not ordinary agile delivery, generic user interviews, research synthesis, feature prioritization, OKRs, positioning, valuation, or permission to release a careless product.

Adjacent concepts: customer development, MVP, concierge test, wizard-of-oz test, smoke test, fake door test, cohort metrics, actionable metrics, innovation accounting, pivot, persevere, value hypothesis, growth hypothesis.

One-line analogy: Lean Startup is instrument flying for product uncertainty: move in small loops, read the gauges, and change course from evidence.

Common misconception: MVP does not mean "smallest thing we can ship." It means the least effort that can produce validated learning about a specific assumption.

Lean Startup

Domain Context

Use Lean Startup when the task involves a new venture, new product, new feature, new business model, innovation program, nonprofit program, or internal initiative where the core uncertainty is whether a customer, user, market, or stakeholder will respond in the expected way. The method is strongest when the team can run small experiments before committing to a full build or scale-up.

Use public, aggregate, or synthetic examples. Do not put personal data, customer identifiers, payment details, private financials, raw interview transcripts, or confidential roadmap details into examples or evals unless the user supplied them and the active task permits that handling.

Lean Startup does not replace judgment. It structures learning. A good output should make clear what is being learned, how evidence will be collected, what threshold will trigger each decision, and why the proposed MVP is the smallest ethical test of the risky assumption.

Coverage

This skill teaches agents to:

1. Decide whether Lean Startup fits the decision context.
2. Separate vision, strategy, assumptions, experiments, and execution work.
3. Name leap-of-faith assumptions, especially value and growth hypotheses.
4. Select the riskiest assumption rather than the easiest assumption to test.
5. Design MVPs and experiments that produce validated learning, not just demos.
6. Choose experiment types such as concierge tests, wizard-of-oz tests, smoke tests, fake-door tests, landing pages, prototypes, pilots, pre-orders, and manual service tests.
7. Define actionable metrics, cohorts, thresholds, and learning milestones before building.
8. Use innovation accounting to track learning progress under uncertainty.
9. Make pivot, persevere, stop, or next-experiment decisions from evidence.
10. Avoid common failures: vanity metrics, post-hoc success criteria, overbuilt MVPs, unethical deception, and scaling before the value or growth hypothesis is supported.

Philosophy of the skill

Lean Startup is useful because it changes the unit of progress. In known execution work, progress can often be measured by completing planned output. In a startup-like environment, output can be perfectly executed and still worthless because the underlying assumptions are wrong. The method therefore asks a different question: what did the team learn that reduces uncertainty about a sustainable product or business?

The build-measure-learn loop is not "build something, launch it, inspect analytics later." The loop starts with a learning question. Building is the cost paid to create an observable customer or stakeholder reaction. Measuring is only useful when the metric can change the next decision. Learning is only validated when the evidence tests the hypothesis that mattered before the result was known.

The word "minimum" in MVP is a constraint on waste, not a license for low craft or user harm. A concierge test, wizard-of-oz test, landing page, manual prototype, or pilot can be more valid than a thin software release when it tests the assumption with less build effort. The right MVP is the smallest ethical intervention that can answer the current learning question.

Workflow

1. Decide Whether Lean Startup Fits

Use Lean Startup when the request is about uncertain demand, business-model viability, adoption, behavior change, pricing, channel, retention, or growth before scale.

Do not use it as the primary method when the user needs:

If the request lacks a hypothesis, ask for or infer one and label the inference.

Vision or opportunity:
Customer / user / stakeholder:
Current belief:
Value hypothesis:
Growth hypothesis:
Riskiest assumption:
What decision this experiment must inform:
Time / budget / ethical constraints:

2. Name the Riskiest Assumptions

List the assumptions that must hold for the plan to work. Then pick the one that combines high uncertainty with high consequence.

Do not spend the first experiment on an assumption that is easy to test but not decision-changing.

3. Convert the Assumption Into an Experiment

Write the experiment before proposing the build.

Hypothesis:
Why this is the riskiest assumption:
MVP / experiment type:
What will be built or simulated:
Who will experience it:
Metric:
Baseline:
Success threshold:
Failure threshold:
Sample / exposure:
Timebox:
Decision rule: pivot / persevere / stop / next experiment
Ethical guardrails:

Choose the MVP form that answers the learning question with the least waste.

4. Define Actionable Metrics Before Building

Use metrics that can change the next decision. Prefer behavior over opinion and cohorts over aggregates.

Vanity metrics are not always big numbers. A small number can still be vanity if it cannot change the decision.

5. Run the Build-Measure-Learn Loop

Answer in this order:

1. Learn: What must be learned next?
2. Measure: What observable evidence will decide the question?
3. Build: What is the smallest ethical thing needed to create that evidence?
4. Run: Expose the right audience under the stated constraints.
5. Interpret: Compare results to the pre-declared thresholds.
6. Decide: Pivot, persevere, stop, or run the next experiment.

The written answer can still present the loop as build-measure-learn, but the agent should design it backward from learning to measurement to build. This prevents overbuilding.

6. Use Innovation Accounting

Regular accounting tells whether an existing business is performing. Innovation accounting tells whether a team is reducing uncertainty in a new business model.

Track:

• assumption being tested
• experiment run
• metric and threshold
• result
• confidence gained or lost
• decision made
• next assumption to test
• cost and time of learning

Use a learning ledger:

7. Decide Pivot, Persevere, Stop, or Next Experiment

Make the decision from evidence, not from effort already spent.

Name the pivot type plainly. Do not use "pivot" as a euphemism for continuing without a learning-based change.

Output Template

Lean Startup validation plan

Decision:
Customer / user:
Vision:
Riskiest assumption:
Hypothesis:
MVP / experiment:
Why this is minimum:
Metric:
Threshold:
Sample / timebox:
Ethical guardrails:
Expected learning:
Decision rule:
Next loop if persevere:
Pivot options if not supported:

Verification

• The plan states the riskiest assumption and why it matters.
• The MVP is tied to a learning question, not just to a smaller release.
• The metric is actionable, behavior-based where possible, and tied to a pre-declared threshold.
• The answer separates value hypothesis, growth hypothesis, and execution work.
• The experiment includes sample, timebox, and decision rule before build effort begins.
• Vanity metrics, compliments, and unsegmented aggregate numbers are not treated as validated learning.
• Ethical guardrails are named for smoke tests, fake-door tests, wizard-of-oz tests, and user-facing experiments.
• Pivot/persevere/stop recommendations cite evidence and avoid sunk-cost reasoning.
• The next loop tests the next riskiest assumption rather than scaling prematurely.

Do NOT Use When