Principles

Principle

AI-native product building is not “how to use Cursor.” It is how to build any product when generation is cheap, judgment is scarce, and tools rename themselves every quarter.

The handbook is organized around decisions, not tools. Tools live in the last section of every chapter and are allowed to go stale.

The five rules

1. Principles before tools — If a chapter only makes sense with today’s model name, it is tooling, not principle.
2. Humans own outcomes — AI proposes; you commit. Especially for architecture, pricing, and what ships.
3. Prototype to learn, code to own — Vibe tools are for learning speed, not for pretending you have a product.
4. Artifacts over vibes — Specs, tests, and trails beat “it felt right in the demo.”
5. Version in public — Ship v0.1. Revise when tools change. A living handbook beats a frozen book.

Principles (years)
↓
Decisions (quarters)
↓
Workflows (months)
↓
Tooling (weeks)

Workflow

Each chapter in this handbook follows the same loop:

1. State the principle (what should still be true in three years).
2. Name the decision (when X, do Y; when not-X, do Z).
3. Sketch a workflow (replaceable steps).
4. Link artifacts (templates, specs, checklists in /templates).

Tooling

Tooling sections are intentionally small and dated. Replace them as the landscape shifts. The decisions above should not.

Common mistakes

• Treating the prototype as the product because demos look finished.
• Letting agents implement architecture you cannot explain in one whiteboard.
• Optimizing for token speed instead of revision speed — fast generation, slow judgment.

Artifacts

• templates/spec-stub.md — one-page spec before agentic implementation.