The 10 limits of vibe coding for a mobile app
Vibe coding is enough for a prototype, not for security, App Store review, in-app purchases, or maintenance. AI-generated code fails security checks in 90% of cases.
Article summary
- A study published in 2025 concludes that AI-generated code fails security checks in 90% of cases.
- On Google Play, 31% of churn is involuntary, caused by billing failures that vibe coding does not configure.
- 77% of users abandon an app in the first 3 days. An onboarding that is only functional does not address that funnel.
- Six months after launch, an outside developer takes 2 to 4 weeks to understand an undocumented architecture, at €500 per day.
- Rebuilds of vibe-coded projects documented by firstapp take 2 to 6 weeks to reach a production level.
Summarize this article with
Vibe coding removed a real barrier: having to know how to code in order to ship a prototype. It did not remove the technical constraints of mobile development in production. firstapp documents here the 10 concrete limits of vibe coding, with the data and incidents that illustrate them. The goal is not to discourage. It is to avoid bad surprises 3 months from launch.
1. Guaranteeing the security of authentication flows
This is the most documented limit, and the most expensive.
Vibe coding tools generate functional code, meaning code that works in normal cases. They do not generate code that resists adversarial cases: an attacker testing combinations of credentials, looking for exposed endpoints, or exploiting a bad permissions configuration.
A real incident documented in 2025 on the Lovable platform exposed flaws that made it possible to access private applications using only their public identifier, with no authentication at all. The flaw bypassed the login system completely. The same platform also had problems with passwords stored in plain text and XSS vulnerabilities.
According to a study published in 2025, AI-generated code fails security checks in 90% of cases, even when it is functionally correct. This is not a problem of intent. It is a problem of method: security requires reasoning about adversarial cases, not only about normal use cases.
| Common vulnerability type | Description |
|---|---|
| Exposed endpoints | APIs accessible without valid authentication |
| Hardcoded credentials | API keys written directly into the source code |
| Poor input validation | Susceptibility to SQL injection or XSS |
| Excessive permissions | The app asks for more access than it needs |
| Non-expiring tokens | User sessions that never end |
Newsletter
Once a week.
Last issue
We opened 10 apps to see how they sell a subscription.
+534 people read us each week
2. Passing App Store review on the first try
Apple regularly rejects apps for reasons that have nothing to do with code quality: missing legal links, prices displayed badly on the paywall, a missing "Restore Purchases" button, or an interface judged misleading.
The problem with vibe coding is that these rules are not encoded in the tools' default prompts. The AI generates a functional interface, not necessarily an interface that complies with Apple's Human Interface Guidelines.
The most frequent App Store rejection reasons on vibe-coded apps:
| Rejection reason | Apple guideline concerned |
|---|---|
| Privacy policy missing or unreachable | Guideline 5.1.1 |
| Price or billing period not displayed | Guideline 3.1.1 |
| Restore Purchases button missing | Guideline 3.1.1 |
| Test account not provided to the reviewer | Guideline 2.1 |
| UI judged misleading (dark patterns) | Guideline 3.1.1 |
| Dynamic code execution after review | Guideline 2.5.2 |
Apple intensified its checks on AI-generated apps in 2025, specifically targeting apps that execute or generate dynamic code after review, which is incompatible with the self-containment policy for iOS apps.
A review rejection can take 2 to 7 business days. On a project with a deadline, every rejection costs time and money.
3. Integrating in-app purchases correctly
Integrating iOS and Android subscriptions is one of the most complex topics in mobile development. It is not a technical integration in the classic sense. It is a system with states, webhooks, edge cases, and rules specific to each platform.
What vibe coding typically generates: a call to the purchase API, a confirmation to the user, and access to premium features.
What it systematically forgets:
- Handling renewals and expirations
- Server-side renewal webhooks (Server Notifications)
- The billing retry case when a payment fails
- Reconciling states between the App Store and Google Play
- Handling Restore Purchases for users who reinstall
- Cross-platform entitlements (the same subscription on iOS and Android)
On Google Play, 31% of churn is called "involuntary," meaning it is caused by unrecovered billing failures, according to the RevenueCat 2026 report. That churn is almost entirely avoidable with a proper RevenueCat or Adapty implementation. Vibe-coded code without these tools leaves a permanent leak in revenue.
4. Maintaining the code over 12 months
Code generated by vibe coding is rarely documented, rarely tested, and rarely structured to be modified by someone who did not take part in creating it.
What that means in practice: 6 months after launch, when you want to add a feature or fix a bug, an outside developer takes 2 to 4 weeks to understand the architecture before they can touch anything. At €500/day, that is €5,000 to €10,000 of understanding before a single line has been written.
The signals of a vibe-coded app that is hard to maintain:
| Signal | Consequence |
|---|---|
| No automated tests | Every change risks breaking another feature |
| Undocumented architecture | Very high cost of onboarding a new developer |
| Packages hallucinated by the AI | Dependencies that do not exist or are unmaintained |
| No separation of responsibilities | Impossible to change one component without breaking everything |
| No error handling | Silent bugs never surface |
5. Managing involuntary churn (billing)
Distinct from voluntary churn (the user who chooses to cancel), involuntary churn happens when a payment fails and the subscription expires without either the app or the user reacting.
The RevenueCat 2026 figures are clear:
| Platform | Share of churn caused by billing failures |
|---|---|
| Google Play | 31% |
| App Store | 14% |
On Android, nearly a third of cancellations are not user decisions. They are unrecovered billing bugs. An app correctly configured with RevenueCat implements the billing grace period, billing retry notifications, and dunning flows that recover part of this churn.
Vibe coding does not configure these mechanisms. They are not in the default prompts, and they are not visible in the interface until the problems arrive.
6. Optimizing onboarding for conversion
A mobile app's onboarding is not a flow of screens. It is a conversion funnel that decides whether the user stays or leaves in the first 72 hours. 77% of users abandon an app in the first 3 days, according to market benchmarks.
Vibe coding generates an onboarding that works, not an onboarding that converts. The difference:
| "Functional" onboarding | Optimized onboarding |
|---|---|
| Asks for the email first | Shows the value before asking for anything |
| Explains the features | Delivers the first "aha moment" |
| Asks for permissions as soon as the app opens | Asks for each permission at the moment it becomes useful |
| Shows the paywall after signup | Shows the paywall after the first user success |
| No analytics tracking | Every step tracked to identify drop-offs |
A badly designed onboarding does not show up in the code. It shows up in the metrics at D+3, when 80% of users have already left the app. Onboarding optimization is one of the most profitable growth levers available, and it is a skill distinct from development.
7. Architecting for scalability
A prototype that works with 10 users does not necessarily work with 10,000. A mobile app's scalability rests on architecture choices made at the moment of initial development: database structure, cache strategy, state management, separation of services.
Vibe coding optimizes for speed of generation, not for scalability over time. Code generated without an architecture constraint often produces:
- Unoptimized database queries (N+1 queries)
- Business logic mixed into the display layer
- Poorly managed global state that slows the app as data grows
- No pagination on long lists
- Unoptimized images loaded at full size
These problems are invisible at launch. They become critical as soon as the app grows.
8. Managing several developers on the same codebase
Most vibe-coded projects are started by a single person. As soon as the project grows and needs several developers, the problems begin.
Without architecture conventions, without tests, without documentation, a vibe-coded codebase is almost impossible to share. Two developers modifying the same code without a shared structure create conflicts and regressions constantly.
The standard practices missing from vibe-coded code:
- No CI/CD configured (continuous integration and automated deployment)
- No code review (pull requests, code review)
- No consistent naming conventions
- No clear separation between development and production environments
- No secrets management (API keys stored in accessible environment variables)
9. Configuring production monitoring and alerts
A production bug on a vibe-coded app is often discovered by a user before it is discovered by the team. The reason: monitoring (logs, metrics, alerts) is rarely configured in AI-generated projects.
What is systematically missing:
| Monitoring component | Use |
|---|---|
| Structured server-side logs | Tracing errors and abnormal behavior |
| Crash reporting (e.g. Sentry) | Being alerted as soon as a crash happens |
| Behavioral analytics (e.g. Amplitude) | Understanding how users navigate |
| Alerts on key metrics | Detecting a drop in conversion or a rise in churn |
| Performance monitoring | Identifying slow screens or long queries |
Without monitoring, it is impossible to know whether the app is working well or badly after launch. Problems accumulate silently.
10. Localizing the app (price, language, stores)
Localizing a mobile app is much more than translation. It is adapting prices by country, store listings by language, and the app's content to local conventions.
What vibe coding does not generate:
- Pricing localization: Swiss users spend markedly more than the global median, and prices in CHF have to be calibrated differently (according to Adapty 2026 data, median 12-month LTV in Qatar reaches $27.5 versus $19.9 in the United States)
- Apple Custom Product Pages: up to 70 product pages customized by traffic source, now available on the App Store
- Store metadata in several languages
- Date, currency, and address formats by country
Localization is a growth lever underexploited by most French apps. It cannot be improvised in a prompt.
FAQ on the limits of vibe coding
Can you launch a real mobile app with vibe coding alone?
You can launch a functional prototype with vibe coding alone. Launching a production app with real users, payments, and personal data is a different question. The risks around security, App Store compliance, and technical debt are real and documented. The recommendation is to use vibe coding to validate an idea, then move to structured development for production.
Which vibe coding tools are best suited to a mobile app?
For a React Native mobile app, Cursor (an editor with AI assistance) and Claude Code (a terminal) are the best-suited tools because they assume an existing codebase that you improve, which preserves the architecture. "Full generation" tools such as Lovable or Bolt are better suited to web apps or fast prototypes. a0.dev specializes in React Native but remains limited to simple cases. For an app destined for real production, none of these tools replaces a developer who understands what they are generating.
How long does a rebuild of a vibe-coded project take?
The length of a rebuild depends on the state of the code and on what has to be preserved. In the cases documented by firstapp, vibe-coded projects that arrive for a rebuild need 2 to 6 weeks of work to reach a production quality level. The good news: a founder who arrives with a vibe-coded prototype often has 30 to 50% of the work already done in the form of validated functional specifications.
Will vibe coding improve to the point of replacing developers?
The tools are progressing fast. In 2026, they handle increasingly complex cases. But security, compliance with store guidelines, and optimization of conversion metrics assume an understanding of business context that code-generation models do not yet model reliably. The question is not "can the AI generate this code" but "can the AI guarantee that this code is safe, compliant, and optimized in the specific context of this app." The answer is no in 2026.