The era of "write once, run anywhere" is effectively over, and it didn't die from performance bottlenecks or developer fatigue—it was strangled by LLMs. A recent deep dive by Johannes Treitz, creator of Werkbook, argues that AI agents have collapsed the cost of maintaining separate native codebases for iOS and Android. For two decades, frameworks like PhoneGap, Ionic, and React Native served as a necessary evil to avoid building every app twice. Now, with agents capable of navigating the labyrinthine hell of Xcode certificates and Android provisioning profiles, the abstraction layer has become a liability rather than an asset.

The Death of the Abstraction Tax

Treitz’s experience with Werkbook, an app for blue-collar workers, illustrates the practical shift. Initially, he utilized Hotwire Native to bridge his Ruby on Rails backend with mobile frontends. While it worked, the "seams" were visible: memory crashes on specific devices, offline failures in basements, and the persistent need for platform-specific debugging despite shared code. The turning point came when he handed the API to Claude and asked for a native iOS app using the web app as a blueprint. The agent didn't just generate Swift code; it handled the entire build pipeline, allowing Treitz to ship to TestFlight without ever opening Xcode. The result was a snappier, more robust app that was actually enjoyable to maintain.

Industry Giants Follow Suit

This isn't just an indie developer anecdote; the industry's biggest players are signaling the same move. Shopify, long the most vocal champion of React Native, announced in September that "native is now the future of mobile at Shopify." Citing AI agents as the primary enabler, the company rebuilt its Shop app in Swift and Kotlin in just twelve weeks. Coinbase appears to be following this trajectory, with job postings describing a rewrite of its mobile app from React Native to native "using AI-assisted code generation." When the platforms that built their brand on cross-platform efficiency pivot to native, the framework’s utility as a cost-saver evaporates.

Trusting the Black Box

The skepticism around AI-generated code is valid, particularly for developers who cannot read Swift or Kotlin. Treitz admits he cannot inspect a Swift closure for memory leaks, but he argues that manual code review was always an illusion of security. His own hand-written code had bugs; the difference now is that the AI writes more tests than he ever would, and the security perimeter remains server-side. By treating mobile apps as untrusted clients, the risk of a UI bug exposing sensitive data is mitigated. The developer’s role shifts from reading every line to rigorous device testing and UX tuning, a workflow that prioritizes functional correctness over syntactic familiarity.

Key Takeaways

  • Abstraction Cost vs. AI Cost: The lines of code saved by React Native are no longer worth the performance and debugging overhead when AI can generate native code for free.
  • Native Performance Wins: Treitz reported that the AI-generated native apps were "snappy, robust, polished," solving the specific memory and offline issues that plagued the cross-platform wrapper.
  • Enterprise Validation: Shopify’s 12-week rebuild of the Shop app in native languages, driven by AI, sets a new benchmark for how quickly large codebases can be migrated.
  • Security Model Matters: This approach only works if the backend (e.g., Rails) handles all logic and security, rendering the client-side language irrelevant to data integrity.

The Bottom Line

Cross-platform frameworks were a solution to a labor shortage that AI just fixed; if you’re still maintaining a React Native wrapper for cost reasons, you’re paying a tax on your own obsolescence.