A writer with no coding background shipped an iPhone app in 13 days using Claude Code
A law graduate who writes about tech had an AI write every line of an App Store app, and the hard parts turned out to be review, testing and responsibility.
The short version
- All code for the calendar app came from Claude Code; the developer never read it and only described features and fixes.
- From first concept to first App Store submission took about 73 hours; Apple's review then sent the app back twice.
- Cross-checks with Claude and OpenAI's Codex caught a risky link-handling flaw and a calendar-saving bug.
- Extra spending was the $99 yearly Apple developer fee; existing $200 and $20 monthly AI plans covered the rest.
Why it matters
The bottleneck in building simple software is moving from typing code to deciding what you want, testing it and owning the result. That widens who can build an app, but this case shows it does not remove the need for checking, accessibility work and dealing with a platform's gatekeepers.
A non-engineer on the App Store
As BigGo Finance reported, a writer who covers smartphones and IT published an iPhone app 13 days after starting, with almost no hands-on programming background. The developer studied law at university and had only picked up programming basics while preparing for Japan's Fundamental Information Technology Engineer Examination. They had put web apps online before but had never made a native app, meaning one built specifically for the iPhone.
The app is called Tsuranari Calendar. It is a calendar designed for the inner screen of the foldable iPhone the source calls the iPhone Duo. Work began on the night of October 23, the day that phone went on sale.
That night the developer talked through the concept with Claude and produced a document describing the screens and requirements. By the next night, a version was running on a real device. Three nights after that it went to Apple, about 73 hours after the idea first took shape.
How the AI did the work, and where the human stayed in charge
The developer used what is called vibe coding: the human does not read the code at all. Instead they describe what they want, look at the result on a phone, and say what to fix. The tool was Claude Code, Anthropic's coding assistant, which can read and write files on a computer and run commands to check that things work.
The developer did not give the AI free rein. The starting instructions required Claude Code to explain what a change would affect, and its plan, before touching anything, and to wait for approval. An "auto mode" also used a second AI model to block operations it judged dangerous. In practice it stopped attempts to rewrite the change history on GitHub, a service that stores code.
There was a setup puzzle too. Building iPhone apps needs Apple's Xcode software on a Mac, but the developer normally works on Windows. Using Claude Code's remote-control feature, plus tools called SSH and Tailscale that connect computers securely, the AI on the Windows machine could build and launch the app on a Mac mini by the second day. The developer also had the AI write more than 300 tests and rerun them after every fix.
Apple's review and the bugs nobody wrote on purpose
The part that did not move at AI speed was Apple's review. The first morning after submission, Apple asked for more information because the developer account had little history. The developer explained the app in English and attached a screen recording from a real device. The next reply took nearly five days.
The second rejection concerned the wording on the screen that asks permission to use the calendar. The developer changed a button from "Allow Access" to "Continue" and the app was approved shortly after midnight on the 13th day. Apple aims to review more than 50% of apps within 24 hours and more than 90% within 48 hours, but extra questions slow things down.
Because nobody read the code, the developer had Claude Code check it for safety problems and asked OpenAI's coding AI, Codex, to read it as well. The checks found that addresses starting with "shortcuts://", which launch another app, could sneak in through links pulled from notification emails. The developer limited which link formats are allowed. Another bug could save an event to a different calendar if the target calendar was deleted mid-edit. The app now declines to save and asks the user to choose again.
Accessibility and cost
Accessibility checks turned up more. Event displays broke when text size was set to maximum, and dates were harder to see for some kinds of color vision. The web accessibility guidelines known as WCAG ask for a contrast ratio of 4.5 to 1. iOS's standard red measured 3.55 to 1 even for typical vision and 2.93 to 1 for people with deuteranopia, a form of red-green color blindness. The developer adjusted the red and blue to pass. A separate bug showed only "8" as the year under the Japanese era calendar setting, so the app now uses the Western calendar.
On cost, the developer signed up for nothing new. They used a $200-a-month Claude plan and a $20-a-month ChatGPT plan already paid for work, and added only the $99 annual Apple Developer Program fee. Claude Code usage reached about 1.76 billion tokens by November 8, tokens being the small chunks of text an AI processes. About 1.73 billion were cache reads, which reuse earlier conversation, and roughly 23 million were new input and output. At API prices the source puts that at about ¥86,000, but the flat plans meant no extra charge.
The basic app is free, with no ads and no server of its own. After launch, the developer added a ¥300 color-theme option called Wa no Iro.
What this does and does not prove
The case shows that someone with little programming experience can ship a practical app with these tools. It also shows what stays human: deciding what to build, testing on real devices, handling Apple, and taking responsibility for bugs after launch.
The developer is candid about the limits. AI checking AI is no substitute for an expert audit, and because the AI wrote the tests, its own misunderstandings can leak into them. This is one person's account of one small app, not proof that every project will go so smoothly.
What to watch
- Will Apple's review process change as more non-engineers submit AI-built apps, or will new-account friction persist?
- How well do AI-to-AI code checks hold up against audits by human experts?
Try it
Before asking an AI to build anything, write a short plain-language description of what done looks like, then test the result on a real device and ask a second AI tool to review it for safety issues.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by BigGo Finance · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
