Skip to content

Guides

Vibe Coding an App: A Law Major's 13-Day iPhone Launch

Vibe coding an app got a real test: a law major reportedly shipped a native iPhone app in 13 days. Here's what's confirmed and what isn't.

On this page
  1. What happened: a law major, an App Store listing, 13 days
  2. Timeline and the 73-hour figure
  3. How vibe coding worked in this case
    1. The loop in practice
    2. The ceiling
  4. Where the human still mattered
  5. The review hiccup: "Allow Access" became "Continue"
  6. What to do if you build apps without code
  7. Build it with FromScratch
  8. FAQ
Illustration of a laptop and an iPhone showing a calendar app, representing vibe coding an app
Photo by Ed Hardie on Unsplash

Vibe coding an app used to mean a weekend of hacking for people who already knew how to code. A recent case suggests it can now get a non-engineer much further. Shattered reported that a law major with basic programming knowledge shipped a native iPhone app to Apple's App Store in 13 days, with Anthropic's Claude Code doing much of the work. This article walks through what the report says, separates the confirmed facts from the loose claims, and turns the story into a practical checklist for anyone building software without writing it by hand.

What happened: a law major, an App Store listing, 13 days

Shattered summarizes reports from BigGo Finance and TechRadar. According to those reports, the app is called Tsuranari Calendar, and it is live on Apple's App Store. Its developer studied law rather than computer science. Their programming knowledge came from preparing for Japan's Fundamental Information Technology Engineer Examination, a general IT qualification that covers the basics rather than production app development.

The developer had built web apps before. What they had never done was build a native app, which involves Apple's tooling, testing on a device, and the App Store submission process. Those are the steps where first-timers usually get stuck, and they are the steps where AI help matters most.

Some things are not established. The developer's name is unpublished. The reports contain no on-record quote from the developer or from Anthropic. Nothing confirms how much of the code the AI wrote. The phrase "every line came from AI" is a claim you will see repeated, but the available reporting does not support it. The story is still useful, just in a narrower form: a non-engineer used an AI coding tool to get from an idea to a live app in under two weeks.

Timeline and the 73-hour figure

The reported sequence is short:

a computer screen with a calendar on it
Photo by Ed Hardie on Unsplash
Step What the reports describe
October 23 Development begins, the same day the iPhone Duo is reported to have launched
About 73 hours in First submission to Apple, per the reported concept-to-submission window
Three nights later Review submission, as described in the coverage
Day 13 after launch Approval, shortly after midnight

The 73-hour number is the most eye-catching figure, and it needs a caveat. The reports describe the review submission as "three nights later," while also putting the first submission about 73 hours after the concept. Those framings are close but not identical, and the sources do not reconcile them. Treat 73 hours as an approximate figure.

The total of 13 days also covers more than coding. It includes device testing, waiting for review, a wording change, and resubmission. If you plan your own project, budget time for that whole loop, not only for the build. For a first native app, the review cycle is often the slowest part.

How vibe coding worked in this case

The coverage defines vibe coding as a method in which the human avoids reading the generated code. The developer describes what they want, the AI writes it, and the developer judges the result by running the app. The code stays closed, like a sealed box the developer tests from the outside.

The loop in practice

The working cycle looks something like this:

  1. Describe the next change in plain language.
  2. The AI revises the app.
  3. Install the new build on a device.
  4. Tap through the screens and note what looks wrong.
  5. Report the problem back, and repeat.

That loop lets a non-engineer move quickly. The developer acts as product owner and tester, not as debugger. It is also why a working version could exist within a few days of starting.

The ceiling

The approach has an obvious limit. If you never read the code, you cannot judge its quality, its security, or how well it will hold up as the app grows. You can only judge behavior: does the screen show the right events, does the button do the right thing. For a small calendar app, behavior may be most of what matters. For anything that handles payments, health data or customer records, behavior is only the starting point. We cover that risk in more detail in our piece on vibe coding security.

Where the human still mattered

The most important point in the story is not what the AI did. It is what the human kept. The developer still made every product decision: what to build, what counted as finished, and when to submit to Apple. Claude Code could draft and revise software, but it did not decide which app deserved to exist.

rose gold iPhone 6s
Photo by Jessica Lewis 🦋 thepaintedsquare on Unsplash

The developer also handled the App Store review response, which meant deciding how to change the app's wording. That is a judgment call about how the product talks to its users, and it was not something the tool could settle alone.

So the accurate reading is not that AI replaced the developer. It is that the tool shrank the skill barrier between an idea and a working native app, while judgment, testing and accountability stayed with the person. Whoever ships the app owns the result, whether or not they wrote a line of it.

The review hiccup: "Allow Access" became "Continue"

The smallest detail may be the most useful. Before resubmitting, the developer changed a button label from "Allow Access" to "Continue." Approval followed shortly after midnight on the 13th day.

The reports do not quote Apple's reviewer, and they do not explain why the label mattered. The sequence suggests the first submission did not clear review as written, but the sources do not say that outright, so treat the cause as unconfirmed. Apple's review does look closely at how apps request permissions, and a button that leads into a permission prompt is a common source of friction. That is a reasonable lesson to take, even without a confirmed cause.

The practical takeaway: wording on screens that ask for access to location, notifications, contacts or calendars deserves the same care as the features themselves. Write it in plain language that says exactly what the user gets and why. Expect at least one review round, and plan for it.

What to do if you build apps without code

Whether you use this specific tool or another, the same habits help.

Write a one-page spec first. Name the user, the three core actions they need, and what "done" means. A vague request produces a vague app, and a spec gives you a fixed test to check against. Our step-by-step guide for beginners walks through this process.

Test every build where users will use it. If your audience is on iPhones, test on an iPhone. If they will open a browser, test in a mobile browser. Log each bug in plain language, with what you did, what you expected, and what happened. Do not try to diagnose the code yourself.

Review security and privacy before real data goes in. If you have not read the code, you need other checks: who can see what, where passwords are stored, and what happens when someone tries to access another person's records. Be especially careful with payment data, personal details and anything regulated.

Keep versions so you can roll back. A change that breaks the app should be easy to undo. Most AI builders save a history of changes, and you should use it. Make a note of which version worked before you try something bold.

Start with a narrow app. A single-purpose tool with a few screens is easier to test and easier to explain to a reviewer than a sprawling product. A study tool or a personal calendar is a sensible first project, which is why we wrote a guide on building a study app without coding.

Build it with FromScratch

FromScratch is an AI app builder. It builds web apps with a React front-end, a Python (FastAPI) back-end and a database. It does not build native iOS apps. If your goal is an App Store listing, this is the wrong tool for that job. If your goal is a calendar your team or customers can open in a browser, here is how to build one.

  1. Create a project and describe the app. Open the chat and type the description. You do not need to name a framework or a database; the agent chooses them.
  2. Wait for the live preview. The app appears in a preview within minutes. Open it on your phone and check the layout first.
  3. Refine by chatting. Ask for one change at a time, such as "make the week view show seven columns" or "add a delete button to each event." Check each change before asking for the next.
  4. Add login and test it with two accounts. Create two users and confirm each sees only their own events.
  5. Restore an earlier version if something breaks. Every change is saved, so you can roll back rather than rebuild.
  6. Publish to a public URL or your own domain when you are satisfied.

Paste this prompt into the chat to start:

Build a web-based calendar app where I can add events with a title, date, time, location and color, and switch between month and week views. Add email login so each person sees only their own events. Use a clean layout that works well in a mobile browser, and save every event to a database so it persists between sessions.

After the first version runs, test the login with two accounts before you add anything else. If you want more context on the process, see our AI app builder overview.

FAQ

Can someone with no coding background build an app with AI?

Yes, for certain kinds of apps. The Shattered case involved a developer with basic programming knowledge, not zero experience, and that is worth remembering. Non-engineers can build small, well-defined tools, but they still need to define the product, test it carefully, and take responsibility for what they ship.

Does vibe coding produce a finished App Store app?

Not on its own. The reports describe the App Store submission as a separate process that includes review, possible rejection, and resubmission. The AI can help produce the app, but the developer still has to complete the native submission steps and respond to Apple's feedback.

Is the Shattered story fully verified?

No. The reporting does not name the developer, includes no on-record quotes, and does not confirm how much of the code the AI wrote. The 73-hour figure is approximate, and the cause of the review issue is not confirmed. Treat the story as a useful case study, not a benchmark.

Can FromScratch build a calendar-style app for the browser?

Yes. FromScratch builds web apps with login and a database, so a calendar that stores events per user is a good fit. You can start from the prompt in the section above and refine it through the chat.

Written by

FromScratch TeamEditorial team

The team building FromScratch, the AI app builder that turns a plain-language description into a real, published full-stack app.