Skip to content

Examples overview

Six worked examples — one per profile — each a fictional product taken end-to-end through the Stage-1 to Stage-3 pipeline. Each shows the Context Profile fill, the resolved profile, the 10-12-phase plan an AI tool produced from the profile-filtered Mandatory items, and the delta-bilanz against a vanilla "build me a SaaS" prompt.

If you read only one, read B2B SaaS — HiringDesk. It's the densest case (no V1 gaps to flag, full 264-item profile activation) and the Step 5 delta-bilanz is the cleanest demonstration of "what vibecodingisold adds that a vanilla AI plan would have skipped."

Index

Profile Example Fictional product Profile-specific lens
b2b-saas-web b2b-saas-example.md HiringDesk — multi-tenant ATS for HR teams, Stripe billing, GDPR, internal LLM resume-parsing Densest case, no V1 gaps to flag
b2c-saas-web b2c-saas-web-example.md PantryPal — consumer recipe organizer with pantry tracking, freemium, Stripe consumer billing Anti-abuse cluster + CDN/WAF + RUM as elevated concerns
marketplace marketplace-example.md BoardSwap — peer-to-peer board game rental marketplace, Stripe Connect with escrow 4 V1 gaps honestly flagged: trust & safety, escrow state machine, two-sided rating, search ranking
ai-wrapper ai-wrapper-example.md CaseLens — legal-brief summarization copilot, RAG over CourtListener + firm briefs 5 V1 gaps honestly flagged: cost monitoring, RAG security, output filters, eval harness, prompt registry
mobile-app mobile-app-example.md ShelfTrack — B2C mobile inventory app, React Native, on-device LLM recipe suggestion 5 V1 gaps honestly flagged: push, app-store, on-device encryption, deep links, cert pinning
internal-tool internal-tool-example.md OpsConsole — internal admin console for fictional 250-employee SaaS Lattica, SOC 2 in scope "Internal does NOT mean unsafe": 184 items, 8 explicit operational skips with ADR-numbers, security stays full

How to read each example

Every example follows the same Step 1 — Step 7 structure:

  1. Step 1 — Filling the Context Profile. The YAML block at the top of PLAN-REQUIREMENTS.md that anchors the rest. Mission, target users, stack, scale, compliance, multi-tenancy, region posture.
  2. Step 2 — Activating the profile. Which Mandatory items light up, which Conditionals trigger and why, which are skipped/deferred and what the documented re-trigger condition is.
  3. Step 2b — Honest about V1 gaps (only for marketplace, ai-wrapper, mobile-app — profiles with documented gaps). What V1 doesn't yet cover for this product type, and what to fill in by hand until V2.
  4. Step 3 — Feeding the AI tool. The slash-command sequence (or the prompt template paste-flow) that hands the artifact to the planner.
  5. Step 4 — The plan AI produced. A 10-12-phase plan dense with item cross-references in (role.md Item N) or (**role N**) format. Every reference is verified to point at an actual item.
  6. Step 5 — What's missing. Delta bilanz: vanilla AI plan vs vibecodingisold-augmented plan, with item counts per category. The headline number for HiringDesk is ~110 items present in the augmented plan that the vanilla plan would have missed.
  7. Step 6 — What the user still has to fill in. Vendor choices, SLO numbers, domain modeling, UI/UX, marketing — explicitly out of scope for the Stage-2 layer. These are the founder-decisions that no tool can make for you.
  8. Step 7 — What V1.1 should add. Dogfooding-driven candidates specific to that example's product type. Useful as input for the V1.1 backlog.

Why fictional products

The examples teach the pattern, not a specific stack. Real products have specific business models, specific compliance regimes, specific scale constraints — none of which generalize. Fictional products let us pin the example to a plausible scenario without committing to a particular real company's choices.

Each fictional product is realistic enough that you can map your own product onto it:

  • HiringDesk is structurally any B2B SaaS with Stripe + GDPR + per-tenant data.
  • CaseLens is structurally any RAG product with multi-tenant document isolation.
  • OpsConsole is structurally any internal tool with production data access.
  • BoardSwap is structurally any marketplace with held funds.
  • ShelfTrack is structurally any consumer mobile app with on-device intelligence.
  • PantryPal is structurally any high-DAU consumer web app on a freemium model.

What these examples are NOT

  • Not a tutorial. They show the output, not the keystrokes. For setup, see Quick Start.
  • Not a substitute for reading the role files. They illustrate the items in context, but the canonical specification lives in core/roles/<role>.md on GitHub.
  • Not exhaustive. Each example references 80-140 items. The full 314 are in the role files.
  • Not endorsements of specific vendors. When an example picks Hetzner, Stripe, Anthropic, Okta, etc. — those are illustrative concrete choices, not recommendations. Vendor selection is in Step 6 of every example as a founder-decision.

Reading order recommendation

If you have 30 minutes:

  1. Read HiringDesk (B2B SaaS) end-to-end. Best signal-to-noise.
  2. Skim OpsConsole (Internal Tool) Steps 1–2 to see how a profile trims (not just filters).
  3. Skim CaseLens (AI Wrapper) Step 2b to see what "honest V1 gap" means in practice.

If you have 5 minutes:

  1. Read HiringDesk Step 5 (delta bilanz) only. That's the value pitch in numerical form.

Where the worked examples live in the repo

All 6 example files are at examples/ in the repo, plus the index examples/README.md which expands the table above with the full structure.

Direct GitHub link: examples/ on 0zoriginals/vibecodingisold.