How to vibe code an app: a step-by-step guide for non-technical founders

Vibe coding is building software by telling an AI what you want in plain words and letting it write the code. You describe a screen, the AI builds it, you try it, and you say what to change. That is how a founder who studied finance, not computer science, built Prayer Lock, an iPhone app, in about three days.

This guide comes from Hook Studio. Hook Studio is an AI reel maker that makes TikTok and Instagram reels from an app's real screens for founders who want to test an idea. So it covers both halves of the job: how to vibe code an app step by step, and how to find out whether anyone wants it.

Vibe coding means you describe what you want, an AI writes the code, and you judge the result by using it instead of reading it. You talk about screens and buttons; the AI deals with the code.

The term comes from Andrej Karpathy, an OpenAI co-founder who later led AI at Tesla. On February 2, 2025, he posted on X about a way of coding where you "fully give in to the vibes, embrace exponentials, and forget that the code even exists." He said it was fine for throwaway weekend projects. In March 2025 Merriam-Webster listed "vibe coding" as slang and trending, and on November 6, 2025, Collins Dictionary named it its Word of the Year.

Vibe coding vs. coding with AI

Developers use AI too, but they read and own every line it writes. Vibe coding is when you do not. For a non-technical founder that is the whole point: you can try an app idea without learning to program first. It is also the risk, and we cover it below.

Vibe coding is good for small apps that do one job and for first versions you plan to change. It is a poor fit for anything where a bug can cost someone money or expose their private data.

• A one-job phone app. Prayer Lock does one thing: it blocks apps like Instagram until you pray, then unlocks them.
• A web app or a waitlist page that explains the idea and collects emails.
• A clickable demo you can screen-record for a video before the real app exists.
• A tool for yourself or your team, like a tracker or a dashboard.

Leave these for later, or for a developer: apps that hold payments, health records, or other people's personal data. If you are not sure how small your first version should be, start with what an MVP is.

Keep the AI's job small at every step: one category that already pays, one screen per ask, and no login or database in the first version. Four of these steps come from how Mau Baron built Prayer Lock; the others keep the build fixable and get it into the stores.

1. 1Pick a category where people already pay. Look at the top apps in your category on the App Store. Mau did not invent a category: he found Pray Screen, an app that already locked phones for prayer and already had users, and set out to make a better one.
2. 2Collect their screens. Download the apps that earn and screenshot the onboarding, the paywall, and the main screen. Mau put the onboarding screens of the top apps in Canva before he wrote any code. Study them; do not copy them.
3. 3Write the app's one job on one page. What it does in one sentence, the screens in order, and what it saves. Paste that page into the tool at the start of every session.
4. 4Build one block at a time. Ask for one screen, try it, then ask for the next. Mau built screen by screen instead of asking the AI for the whole app at once. Small asks are easier for the AI to get right and easier for you to undo.
5. 5No login and no database in v1. Keep the data on the phone. Prayer Lock launched this way. It means less to build, no user data on a server to protect, and no sign-up screen between the download and the paywall.
6. 6Save a version before each change. Use the tool's history or Git. When a new ask breaks an old screen, go back instead of arguing with the AI.
7. 7Test on a real phone, then submit. Check every button, every link, and what happens offline. Then send it to Apple or Google (fees and rules below).

Apple's rules back up step 5. As of October 2026, its review guidelines say that an app without significant account-based features should let people use it without a login, and that any app that lets people create an account must also let them delete it inside the app. Skipping the login skips both.

To publish, the Apple Developer Program costs $99 a year. Google Play charges a one-time $25, and a new personal account must run a closed test with at least 12 testers for 14 days before the app can go public. Apple says it reviews 90% of submissions in under 48 hours. For Prayer Lock, the coding took about three days and App Store review took longer than that, so plan for a round or two.

If you have never coded, start with a builder that runs in the browser and hosts the app for you. If you want to own the code and keep growing it, use a coding agent. The table sums up what each company says its tool does, as of October 2026. Hook Studio is not affiliated with any of these companies and earns nothing if you sign up.

Vibe coding for beginners: which one to pick

• You want a web app or a waitlist page: Lovable, Bolt, Replit, or v0.
• You want an iPhone app in the App Store: Rork, Vibecode, or Replit (its phone apps need a paid Replit plan).
• You want to own and grow the code: Cursor, Claude Code, or Codex. Mau built Prayer Lock with Cursor and Claude Sonnet, used the Codex command-line tool for bigger changes, and asked the ChatGPT app quick questions.

Pick one and stay with it for the first version. Prices change often, so check each tool's pricing page before you start.

Vibe coding gets you a working first version fast. It does not make the app safe, easy to maintain, or approved. Three limits matter most.

Security

The AI writes code that works, not code that is safe, and you cannot check what you do not read. In 2025, a security researcher found that 170 of 1,645 web apps built with Lovable had a flaw that let anyone reach their users' personal data, because the database had no working access rules (CVE-2025-48757). Skipping the login and the database in v1 removes most of that risk. Before you store personal data or take payments, pay a developer to review the app.

Maintenance

Every new ask can break something that worked, and then you have to describe a bug in code you have never read. In July 2025, Replit's agent deleted a live database during a code freeze in an app that SaaStr founder Jason Lemkin was building, after he had told it not to make changes. Replit added safeguards afterwards. Keep backups, and save a version before every change.

App Store review

Apple reviews the app, not how you built it, so a vibe-coded app fails the same rules as any other. As of October 2026, these are the ones to plan for:

• Unfinished apps. Apple says more than 40% of unresolved issues fall under guideline 2.1, App Completeness: crashes, placeholder content, and broken links. Every app also needs a privacy policy link and a support link.
• A website in a wrapper. Guideline 4.2 asks for an app that goes beyond a repackaged website, and Apple lists web content in an app shell among the common reasons apps fail review. Keep that in mind if your builder only makes web apps.
• Copies. Since June 2026, guideline 4.3(b) says not to submit apps that are indistinguishable from what is already widely available. Study the apps that earn, then make yours meaningfully different.

When an app takes days and an AI subscription to build, the code stops being what separates you from revenue. What is scarce is people who see the app. Everyone is making apps; almost no one tests them with social content first.

Mau Baron's path shows it. Before Prayer Lock he spent years, in his words, building stuff nobody wanted. Prayer Lock's code took three days. What came after was content: up to 40 posts a day across 12 accounts, testing format after format. The numbers from how Prayer Lock hit $100K a month are the best reference we have:

• About 40 posts on one account to find a hook that held.
• The format that won: a person reacting, cut with a screen recording of the app.
• About 10K views a day on one account brought his first $1K a month, with no ads.
• Every 10K daily views brought roughly 200 to 300 downloads a day.

Post a demo before the app exists, or while you build it, and let views, saves, comments, and link clicks tell you whether to keep going. A few short videos take less time than a week of building, and they answer the one question the build cannot: do strangers care?

1. 1Make something to show. The first screen you vibe code, or a clickable mockup made from the screens you collected. Record it on your phone.
2. 2Use the reaction format. A person reacting in the first seconds, then the screen recording, then one line on what the app does.
3. 3Post many hooks and change one thing at a time: the opening line, the first shot, or the reaction. Keep the rest the same so you know what moved the numbers.
4. 4Put a link in your bio: a waitlist page, a TestFlight link, or the App Store link once you are live.
5. 5Make more of the winner. When one hook works, post variations of it instead of starting over.

Prayer Lock's numbers are a reference, not a promise. Another category, account, and hook will get other numbers. What carries over is the method: many cheap tests, then more of what worked. For openings to try, see our viral hook examples. For a founder who used faceless carousels instead of videos, read about the 18-year-old vibe coder and his accountability app.

The expensive part of this test is volume: 40 posts need someone on camera and someone to edit. Hook Studio makes the reaction-plus-screens reel without filming anyone. You describe the reaction and upload your screen recording, and the agent builds the reel in a chat.

• Only the hook is AI video. A still of a person reacting costs 11 credits, and a 5 second Kling 3 clip made from it costs 96. A 5 second hook is 107 credits, about $0.93 on the Concept plan.
• The rest is your real app: your screen recording, on-screen text, and an end card, rendered for a few credits.
• Testing 40 hooks costs about 4,300 credits, around $37 on the Concept plan. A new line of text over the same clip costs only the render, the cheapest test there is.
• The same face in every version. Save an avatar, or star in the reels yourself from a few photos.
• You see the price first. The agent shows each job's price before it runs, and a failed clip refunds its credits. These prices leave out the automatic realism checks, a few credits each.

What it does not do: it does not post for you yet or schedule posts (you download the bundle and post in TikTok or Instagram), and it cannot promise views. A real creator may still convert better than an AI person, and Mau's winners were real people. Use cheap AI reels to find the hook, then pay creators or ads behind the winner, and turn on the platform's AI label when you post. More on the AI reel maker. If you vibe code in Claude, Cursor, or Codex, Hook Studio also works as an MCP server, so you can ask for the reel from the same tool.

Learning how to vibe code is the easy half now. Pick a category that already pays, build one screen at a time, and skip the login. Then, before you spend a month polishing, post the demo and count the people who ask for the link.