Coding agents such as Claude Code, Codex and Cursor can write most of the code for native Swift and Kotlin apps or cross-platform React Native and Flutter apps, but mobile adds steps they cannot fully own: Xcode and Gradle builds, simulators, code signing, store accounts, privacy declarations and review. The fastest results come from choosing a stack the agent can build and test from the command line, and planning the publishing process from day one.
On this page
Agentic coding tools are very effective for web projects because everything they need runs in a terminal: install, build, test, serve. Mobile development is harder. Apps compile with platform toolchains, run on simulators or devices, need signing certificates and must pass human review by Apple and Google. Agents can still do a lot of the work, but only if you set up the project so they can build and verify it, and if you plan the parts that remain manual. This guide explains which mobile paths suit coding agents, where they typically fail and how to get from repository to store listing.
Four paths to a mobile app with an agent
Teams usually pick one of four approaches. Native means Swift and SwiftUI for iOS and Kotlin with Jetpack Compose for Android, which gives the best platform integration but means two codebases. React Native with Expo uses TypeScript, which agents handle very well, and offers cloud builds. Flutter uses Dart and one codebase for both platforms, with a strong command-line toolchain. Web app plus native wrapper keeps the product as a web app, often built with an AI builder or agent, and wraps it in a native shell that adds push notifications and device features.
How well agents handle each path
| Path | Agent strengths | Typical friction |
|---|---|---|
| Native Swift and Kotlin | UI code, view models, networking, unit tests | Xcode project files, signing, simulator interaction, macOS required for iOS builds |
| React Native with Expo | Shared TypeScript logic, screens, API layer | Native module versions, build configuration, upgrade cycles |
| Flutter | Widgets, state management, tests via CLI | Platform channels, plugin compatibility |
| Web app plus wrapper | Everything in the web layer, fastest iteration | Must add real app value to pass Apple Guideline 4.2 |
Making the project agent-friendly
The single most important factor is whether the agent can build and test without a human clicking in an IDE. For iOS, that means the agent can run xcodebuild for a simulator target and run tests from the command line on a Mac. For Android, Gradle tasks for building and unit tests should run in the terminal. Document these commands in CLAUDE.md or AGENTS.md together with the minimum OS versions, the architecture pattern and naming conventions. Keep generated project files under version control and ask the agent to avoid editing them by hand where a tool such as XcodeGen or Gradle configuration can do it reliably. Our guides on building a SwiftUI app with Claude Code and Claude Code with Flutter go deeper on each stack.
Where agents still fail on mobile
Agents frequently produce code that compiles but behaves badly on real devices: layouts that break on small screens, missing permission prompts, background tasks the OS kills, or push notification code without correct entitlements. They also tend to invent API details for platform frameworks that changed recently. Code signing, provisioning profiles and certificates are another weak point, because they live in developer accounts the agent cannot and should not access with broad credentials. Plan device testing, ideally through TestFlight for iOS and internal testing tracks for Google Play, as a fixed part of each iteration.
From repository to store
- Register developer accounts early: the Apple Developer Program costs USD 99 per year, Google Play charges a one-time USD 25 registration fee.
- If you publish under a new personal Google Play account, plan the closed test that is currently required: at least 12 testers for 14 days before production access.
- Prepare privacy information: Apple's privacy nutrition labels and privacy manifest, and Google's Data safety form, matching what your app and its SDKs actually collect.
- Create store listings with screenshots for the required device sizes, descriptions, support URL and privacy policy.
- Upload builds through Xcode or App Store Connect and the Play Console, and submit for review with test credentials if login is required.
- Respond to review feedback quickly; most rejections are fixable with clear communication.
Shortcuts that work
If your product already exists as a web app, wrapping it is often the most economical way to get into the stores, provided the app offers real value beyond the website. WebViewGold turns a web app built with Claude Code, Codex or an AI builder into native iOS and Android apps with push notifications, biometric login and in-app purchases. For the publishing work itself, from accounts and listings to privacy forms and rejection handling, AppSubmitter provides hands-on help until your app is live. Before you submit, compare your build against the App Store submission checklist for Claude Code and Codex projects.
Practical advice for tech leads
Pick the stack your team can review, not just the one the agent writes fastest. A cross-platform framework halves the code an agent produces, but only helps if someone on the team understands its native layer when things break. Keep a Mac build machine or a cloud build service in the loop from the first week, and treat signing and store accounts as company assets with proper access control. At OnlineAppCreator we follow this approach when we take mobile products from idea to operating business, and AppSubmitter grew out of the publishing work behind those launches.
Frequently asked questions
Can Claude Code or Codex build an iOS app without a Mac?
They can write the Swift code anywhere, but building, signing and testing an iOS app requires Xcode on macOS or a cloud build service. Plan for one of the two from the start.
Which mobile stack works best with coding agents?
React Native with Expo and Flutter work well because builds and tests run from the command line. Native Swift and Kotlin also work if the build commands are documented for the agent.
Will Apple reject an app built with AI?
Apple does not reject apps because AI wrote the code. It rejects apps that crash, lack functionality, misstate privacy practices or are only a repackaged website, so test thoroughly and add real app value.