Shopify drops React Native for native Swift and Kotlin, crediting AI coding agents

Shopify is reversing one of the most closely watched bets in mobile development. According to dev.to, the company — which spent roughly five years championing React Native and said as recently as January 2025 that the framework had a strong future at Shopify — is rebuilding Shop and its other flagship apps in native Swift and Kotlin, with no shared codebase between the platforms.

The stated rationale

When Shopify adopted React Native around 2020, the problem was that shipping every feature twice, once per platform, consumed too much engineering capacity. As dev.to tells it, the company's new argument is that AI coding agents have dissolved that cost: internal experiments reportedly showed agents could implement a feature on one platform using the other platform's finished code as a reference, handling translation, testing and review with limited human involvement.

To run this at scale, the company built internal tooling called Helix, which splits work into checkpoints that must pass automated tests, clear an adversarial review stage and receive human approval before the next one starts. A CLI layer lets agents exercise business logic without launching a simulator, which had apparently been a real bottleneck. The Shop app reportedly went from working prototype to a fully rebuilt, shipped app in twelve weeks, with AI doing most of the implementation.

Caveats in Shopify's account

The dev.to analysis, written by someone who has publicly defended React Native for years, questions several parts of the story:

• The performance comparisons predate React Native's New Architecture, the overhaul that improves how JavaScript and native code communicate, so the baseline was an older framework than teams run today.
• Shopify was not using Expo, which handles builds, over-the-air updates and developer clients. Without it, teams effectively rebuild a weaker copy of that tooling, a cost the native-versus-cross-platform framing hides.
• The rewrite trimmed roughly one megabyte of app size, which suggests bloat comes mostly from assets, localisation and media rather than from the framework itself.
• Over-the-air updates, arguably React Native's biggest practical advantage, get almost no mention. They let a team patch a bug without app store review, while a large share of real users always sit on outdated native builds; native code has no equivalent.

The criticisms that hold up

The same write-up concedes genuine problems: dependency management through Xcode and CocoaPods is painful in ways that have little to do with React or JavaScript; the framework's approachability attracts a wider range of skill levels, so ordinary React mistakes land harder on mobile; version upgrades have become unreliable enough that a clean reinstall is often quicker; and teams sit downstream of Meta's roadmap instead of using new platform features the day they ship.

A solo port told a different story

A separate experiment cited on dev.to complicates the timeline. One developer, working part-time, gave an agent a simulator, an existing React Native app and a reference implementation, and got a near-complete SwiftUI port in hours rather than weeks. The likeliest explanation offered is model choice, since large cross-language ports apparently succeed or slog depending on which model does the work.

That experiment also surfaced a point beyond framework speed: SwiftUI's navigation stack recognises a back-swipe only from a narrow strip at the left screen edge, yet many users swipe back from mid-screen, and Apple's own first-party apps ship custom gesture handling to cover both. Native is not an automatic UX win — deliberate work is required either way.

The talent question behind the framework

The analysis argues the most consequential part of this story concerns who maintains React Native. At its peak, Meta reportedly had roughly ten times as many engineers on React Native as on core React and the web side combined. That investment is shrinking: several senior engineers have left Meta over the past year, a number of them joining companies that build AI coding tools. The framework is not going away, but the depth of talent behind it has visibly changed, and a five-year framework bet should weigh maintainer health alongside benchmark results.

Why it matters

If AI agents genuinely make it cheap to build the same feature twice, the economic case for cross-platform frameworks weakens across the industry, and Helix's checkpoint pattern — automated tests, adversarial review and human sign-off at each stage — becomes a template other teams will copy. But the dev.to critique shows the evidence is thinner than the headline: an outdated architecture, absent tooling and an ignored over-the-air advantage all muddy the comparison. For engineering leaders, the practical takeaways are to scrutinise rewrite benchmarks, treat agent-assisted porting as heavily model-dependent, and factor maintainer investment into any long-term framework choice.