AI-Native MVP Stage

Goal

Help founders ship the smallest product that proves the core customer outcome while keeping AI-generated work testable, secure, and maintainable.

Required Inputs

If the founder provides a structured brief, use these inputs:

Validated customer/problem:
MVP outcome to prove:
Current product status:
Technical stack:
Data/security constraints:
Available builders/tools:
Deadline:
Desired output:

Guided Intake

Do not require the founder to know all implementation details upfront. If the request is thin, ask up to five questions first:

1. What customer problem and user outcome has already been validated?
2. What is the smallest workflow the MVP must prove?
3. What exists today: mockup, prototype, manual workflow, or no product?
4. What technical or data constraints matter most?
5. What do you want next: MVP scope, architecture, coding-agent task plan, eval plan, or milestones?

After the user answers, infer reasonable defaults, mark unknowns explicitly, and produce a build recommendation. Do not block on stack details unless the requested output is technical architecture.

Workflow

1. Define the MVP proof target: the user outcome that must become measurably easier, faster, cheaper, or better.
2. Cut scope to the smallest workflow that proves that target.
3. Use references/mvp-scope.md to separate must-have proof from distracting surface area.
4. Use references/technical-architecture.md for architecture, coding-agent guardrails, security, and technical debt prevention.
5. Use references/evals-and-feedback.md to define evaluation, telemetry, bug intake, and customer feedback loops.
6. Return a build plan with milestones, risks, evals, and acceptance criteria.

AI-Native Workflows

Use generic AI roles:

• Coding agent: implement bounded tasks with tests and explicit file ownership.
• Architecture critic: review data flow, security, and maintainability.
• Evaluation assistant: create test cases, golden examples, and failure taxonomies.
• User-research assistant: convert feedback into product decisions.

Exit Criteria

The MVP stage is complete when:

• the core workflow works for real users;
• the product has enough instrumentation to learn;
• the team can distinguish product issues from AI quality issues;
• critical data and security risks are controlled;
• the next launch audience is clear.

Common Failure Modes

• Building a broad product instead of proving one workflow.
• Letting coding agents create unreviewed architecture.
• Shipping AI behavior without evals or regression checks.
• Ignoring data permissions, privacy, and failure recovery.
• Treating demo quality as customer value.

Recommended Outputs

Return the most useful artifact for the request:

• MVP scope brief;
• technical architecture review;
• coding-agent task plan;
• eval and feedback plan;
• milestone roadmap.