The AI kept misreading it
New apps always started as spoken instructions. One day the boss said "make it like Habitica." The AI heard that and built a screen based on whatever it thought 'Habitica-ish' meant. It wasn't the screen the boss had in mind.
What the owner said
Make it like Habitica
So here is what I did
- The AI read 'Habitica-ish' its own way and built a screen from that
- It went in a different direction from what the boss meant — the two sides meant different things by the same words
What the owner didStopped re-explaining after every wrong result and decided to replace words with picking
In words, the AI fills blanks its own way. The fix was to remove the blanks: every name gets a number next to it, and every number carries its own sentence for the AI to read.
So it became numbers plus spec text
The first screen asks for a kind — utility app, game, gamified app, website, or web app (SaaS) — and only the matching questions show up after that. Platform, features, accounts, storage, monetization, layout: every item has a number.
- What people see is a simple name and a picture — "Bottom tab bar," "Character room"
- What the AI gets is an exact sentence — "Fixed bottom tab bar (3–5, icon + label), preserve scroll position per tab" — no vague words
- Everything picked assembles into one document — copy it and paste it straight into the AI
How to use it — three steps
- ① Pick what you're building — utility app / game / gamified app / website / web app
- ② Check off each question — multiple picks are fine, and "Not sure" hands it to the AI's recommendation
- ③ The document appears below — copy it and paste it into a chat with an AI like Claude Code
Every document ends with a "How to work" section telling the AI to settle open questions with the owner before writing any code — so it doesn't just build everything without asking.
Three rebuilds along the way
The tool was first named "Screen Order Form" — pick a screen like ordering off a menu. Once it went public, that name read as a joke rather than a real tool, so it became "Vibe Coding Blueprint."
| Before | Changed to | Why |
|---|---|---|
| Named "Screen Order Form" | "Vibe Coding Blueprint" | It's a public tool anyone can use; the old name read as a joke |
| Examples showed our own project names | Replaced with generic examples nobody would mistake for a real project | A public tool had no reason to carry our own app names |
| Only one choice per question | Multiple choices + a free-text "anything else" box | Most real answers needed several features at once, not one |
These got fixed one by one after launch, by actually using it. Using your own tool is what shows you where it breaks.
People still pick things that don't fit together
For example: pick "stored on device only," then also pick "comments / chat." Comments need multiple people to see the same data, which needs a server — but storage was set to device-only. Nobody can hold every combination in their head.
So 13 conflicting combinations are checked automatically and shown as a warning on screen — "you picked this and that, and they can't both happen" — before any code gets written.
- Storage set to "device only" while also picking a server-dependent feature like comments, chat, or a leaderboard
- No accounts, but a feature that needs to tell users apart (like a leaderboard) is picked
- "Zero cost" priority, alongside real-time multiplayer that costs server money every month
- Target set to "Prototype" (skip saving/login), while saving or login is listed as a must-have
People don't have to catch this by hand anymore.
It used to only show up after the code was already written
Actually used it to design a game
The same day, a defense game idea got run through this tool. Picking "game" as the kind surfaced game-specific questions — screen layout, controls, hero count, reward types. Checking through them produced one document, with a "discuss this first" list attached.
Handing that document to the AI meant it didn't start writing code — it asked through that list first. One document replaced ten rounds of explaining out loud.
This is how it gets used now
Every new app or feature idea starts by opening this tool first. The resulting document is what kicks off the first conversation with the AI. Explaining in words and getting a different picture back has become a lot less common.