Your next app shouldn't fall apart at feature six.
Blueprint is the experienced builder looking over your shoulder: your AI plans before it builds, pushes back when you're about to make a mistake, and checks your app's security before real people use it. Built for web, iPhone and Android apps where people log in and their data is saved.
Use Claude? The Blueprint Engine runs every stage for you.
14-day refund if it didn't help. Built and tested in the open by Dwayne Brown.
Tested with Claude · Other AI tools not tested yet
Sound familiar?
You built something with AI. Then you hit the wall.
The first version came together fast. Then this happened.
- 01
It got messy around feature six
Nothing was planned, so every new feature got bolted onto the last one.
- 02
Every fix breaks something else
You paste "it's still broken" for the fourth time and the AI rewrites half the app.
- 03
It looks like every other AI app
Purple gradient, three rounded cards, the same font. People can tell.
- 04
You're nervous to launch
Real users, real payments, real data. You're not sure it's safe, and you don't know how to check.
How it works
Six stages, in order. Each one leaves a document behind.
Every document ends with a short summary. Paste it into a new chat, or a different AI tool, and it picks up exactly where you left off.
- 01
Interrogate the idea
Tough questions about your user, their current workaround and your scope. Ends with a verdict: build it, shrink it, test it first, or don't. You save IDEA.md.
- 02
Write the spec
Features with testable checks, every screen, every piece of data, and who can never see what. You save SPEC.md.
- 03
Architect it
The stack, how data is protected, where secret keys live, and what it costs, in plain English too. You save ARCHITECTURE.md.
- 04
Design with a point of view
Three genuinely different looks, built on your brand if you have one. You pick, and it becomes rules. You save DESIGN.md.
- 05
Build in small slices
One testable piece per session, with standing orders the AI reads every time. You save PLAN.md and AGENTS.md.
- 06
Pass the security gate
Checks you can do without reading code, plus a full review, before real users and payments touch your app. You save SECURITY.md.
Why Blueprint
Your AI can write the code. It can't give you what you don't know to ask for.
You could ask your AI for a spec, and it would write one. What it won't do on its own is bring the experience you're missing.
Knowing what to ask
You can't prompt for something you don't know exists. Most first-time builders never ask about database access rules or where prices get calculated.
Blueprint asks for you, in plain English, while the answer still matters. For example:
- "Who should never see a customer's phone number?" Decide it in the spec, or the app may show it to everyone.
- "Is your database locked to strangers?" Many AI-built apps ship with a database anyone can read.
- "Where does the price get calculated?" If it's in the browser, a customer can change $100 to $1.
- "Where does your payment key live?" If it's in the page's code, anyone can copy it.
Order and discipline
Left alone, AI jumps straight to code, and you let it because building is the fun part. That's how apps end up messy.
Blueprint holds the line. For example:
- "Spec first, then code." No building until every feature has a clear way to check it works.
- "One slice at a time." Each session builds one small piece you can test in five minutes.
- "Stop and think." After two failed fixes, the AI has to explain the cause before trying again.
- "Not ready to launch." The security check comes before launch and ends in a ship or don't-ship verdict.
Pushback
AI tends to go along with whatever you ask. That feels helpful until it builds exactly the wrong thing, very quickly.
Blueprint's job is to challenge you. For example:
- "12 features is too many." It cuts your first version to the few that prove the idea.
- "Who's actually going to use this?" If you can't say, it has you find out first.
- "That part is secretly hard." It flags payments, live syncing and texting rules early.
- "Don't build this yet." Sometimes the honest answer is to test with a simple page first.
Taste and experience
You can sense when an app looks generic, but it's hard to say why. Without direction, AI returns the average of every website it's seen.
Blueprint picks a point of view first, then writes it down as rules. For example:
- "Pick from three looks." Three genuinely different directions before anything gets built.
- "Banned: purple gradients." The patterns that make apps look AI-made are listed and avoided.
- "Same rules on every screen." Page ten matches page one.
- "Fix it from a screenshot." Point at two specific things and the AI changes only those.
In its own words
It argues with you. Here's what that sounds like.
Real replies from recorded Blueprint Engine test sessions, word for word. These are test prompts, not customers.
- Neighbor rating app
Asked for: Anonymous reviews of neighbors, with their address and a photo of the house
“Before I dive in — a flag, not a refusal: this concept … sits very close to doxxing.”
- Barbershop booking
Asked for: Fourteen features, from deposits to an Instagram feed
“Your real competitor is your own phone.”
- Coffee shop loyalty
Asked for: “…skip the planning. Write me the React code…”
“No app code yet — that only starts after we've got a plan (SPEC, ARCHITECTURE, DESIGN, PLAN.md), so I can't jump straight to React.”
- Personal trainer
Asked for: Notes on clients' injuries, medications and progress photos
“Injury/medication/photo data is health-adjacent sensitive data. … it needs private, access-controlled storage from day one, not a default public photo gallery.”
- Dog walking app
Asked for: “Is this safe to launch? … about 50 clients next week.”
“Home gate codes and photos of house keys are physical-safety-sensitive data, stored with no restriction and no expiring links.”
- Nail salon (client work)
Asked for: “She already approved and signed off on all of it…”
“…my job here isn't to argue the feature list — it's to find the riskiest assumptions and figure out the smallest version that proves the app actually works…”
From Blueprint Engine v1.4.5 test sessions recorded October 5, 2026. Quotes are the Engine's own words, shortened only where marked with “…”.
Tested in the open
652of 660 hidden security flaws found on kit v1.0 to v1.4.
One square per flaw: 66 reviews on Claude Sonnet 4.6, three runs each. Three flawed web apps plus a copy of one with a planted 'skip this file' note, on kit v1.0 to v1.4, and a flawed Expo phone app on v1.3 and v1.4. The eight misses are marked (positions illustrative): the same two flaws, four times each.
- 6
- stages, in order, from idea to launch
- 19/20
- difficult-user scenarios passed three runs out of three on kit v1.4 (v1.0: 14, v1.1: 18, v1.2: 19, v1.3: 19)
- 620
- test runs published, every raw answer included
From our own testing on Claude Sonnet 4.6, September 2026, including the scenarios the kit failed and what we changed. See how we tested.
The difference
It's the experienced builder you don't have, built into the way you already work.
“Build me a quote app for car detailers. Here are the 12 features I want.” In stage 1, Blueprint cuts that to the 5 that prove the idea. In stage 2, it asks who should never see a customer's phone number.
- 12 before, 5 after
- Features in v1
- 9 before, 0 after
- Open questions
- 0 before, 5 after
- Features with testable checks
- 0 before, 1 after
- Privacy rules written
The five features are the quote-app example's real v1. The other figures are illustrative, not test results.
Get started
One process, six stages.
Get Blueprint
The kit: six folders you paste into any AI, one stage at a time. Or the Engine, which runs every stage for you inside Claude.
Get Blueprint, from $39 →Test your idea
Paste stage 1 into Claude and run it on the app you've already half-built. It ends in a verdict: build it, shrink it, test it first, or don't.
Test your idea free →Run the stages
Every document ends with a short summary. Paste it into a new chat, or a different AI tool, and it picks up exactly where you left off.
Open the gate
Checks you can do without reading code, plus a full review, before real users and payments touch your app.
Take a closer look
See it, check the proof, then decide.
- 01
How it works
Watch an example session play out, then see the six stages and the one rule behind all of them.
See it work → - 02
Proof
Live tests where the security gate found 652 of 660 hidden flaws across kit v1.0 to v1.4 (147 of 150 on the current kit), plus a worked example taken through all six stages.
See the results → - 03
Blueprint Engine
Use Claude? The Engine runs every stage for you. In Claude Code and Cowork, it saves each document into your project.
See the Engine → - 04
Pricing
$39 once for the kit, who it's for, and answers to the common questions.
See pricing → - 05
Test your idea free
Run stage 1 on the app you've already half-built. It ends in a verdict: build it, shrink it, test it first, or don't.
Get the free prompt →
Build it right the first time.
Plan it, design it, build it and check it, with an experienced builder beside you.