When to leave the prototype · AI-Native Product Building

Principle

Vibe prototyping tools are learning accelerators, not ownership layers. They help you find shape fast. They punish you if you mistake generated UI for a billable, testable, multi-engineer product.

Most teams switch too early (editing build config while still guessing the problem) or too late (the Lovable app is the product — except it cannot bill, test, or survive a second engineer).

The decision

DEC_003

  Idea
│
▼
Vibe prototype ──(learn)──► still guessing? ──yes──► stay
│                           │
│                           no
│                           ▼
└──────────────────► own codebase + tests + deploy

The fork most founders hit between demo and product.

Workflow

Time-box the prototype (days, not months). Name what you are trying to falsify.
Record the workflow as a checklist, not a screen recording.
When a limitation appears, write it down before switching tools — that note becomes the first engineering ticket.
Migrate vertically: auth → data → payments → polish. Not “rewrite everything because AI said so.”
Keep the prototype URL as a tombstone — what you learned, not what you ship.

Tooling

Lovable, Bolt, v0, Figma Make — same decision framework. Pick whichever gets you to the three exit criteria fastest. None of them are the destination.

Common mistakes

Shipping the prototype because stakeholders saw a polished demo.
Rewriting from scratch without carrying forward the decisions the prototype validated.
Waiting for “one more feature” in no-code because coding feels heavy — that is often fear of ownership, not product sense.

Scenario (composite)

A founder built a client portal in a vibe tool in four days — impressive demo, one design-agency pilot user. After three weeks the pilot ran onboarding eight times (good signal) but hit: no role separation for junior staff, no invoice webhook, export blocked by the tool’s row limits.

They stayed in vibe one more week to confirm the workflow checklist, then migrated vertically: auth first, Postgres schema second, Stripe third. The vibe URL stayed live with a tombstone note. Total migration took twelve days; they did not rewrite screens that were not on the ten-run path.

Artifacts

templates/migration-checklist.md — data model, auth, billing, observability, CI — tick before cutover.

Falsify this

You have zero paying or committed users but “the UI is done” — that is shape learning, not exit criteria.
You have run the workflow twice on the happy path — ten real runs still matter.
You are migrating because coding feels virtuous, not because a hard limitation blocked the next experiment.
The prototype handles billing in a hack — you postponed ownership because migration is scary.
Second engineer onboarding fails immediately — you left too late without tests or deploy docs.