React Native's Peak Moment Undone by AI Agents: The New Cross-Platform Economics

React Native, open-sourced in 2015, has just completed a six-year bottom-up rewrite. Version 0.76 enables the new architecture by default, 0.82 removes the old architecture entirely, Hermes V1 becomes the default engine, and strict TypeScript APIs are now standard. The new architecture—built around JSI, TurboModules, and Fabric—dismantles the inefficient asynchronous Bridge that defined the old era. The State of React Native 2025 survey shows 80% adoption of the new architecture, with 88% of developers believing the framework is moving in the right direction.

From a technical evolution standpoint, this is arguably React Native's most mature and perfect moment. Yet at this exact juncture, Shopify—one of the framework's largest backers—announced its departure. Shopify had spent years migrating its main app (600+ screens) to React Native, long sponsored React Native Skia, and open-sourced high-impact projects like FlashList. As recently as January 2025, Shopify's retrospective emphasized that the transition was "highly successful with a bright future."

In September 2026, Shopify abruptly reversed course: it announced a full return to native development (Swift/Kotlin) and, leveraging AI coding agents, rewrote its consumer-facing Shop app entirely in native code in just 12 weeks. A framework that just finished its foundational rewrite and a major customer that pledged lifelong investment walked in opposite directions at the same crossroads. This is more than a routine technology migration—it signals that the core assumption behind React Native's birth and prosperity has been shattered by AI.

The Ledger Changed, the Formula Must Be Rewritten

React Native's golden era rested on a simple, robust business logic: engineers are expensive, so code should be written once. When Shopify decided to drop Swift and Kotlin for cross-platform in 2020, the logic was pure: avoid two teams duplicating the same logic on two platforms. As engineering director Mustafa Ali put it, "two codebases = 2x headcount = doubled communication and maintenance costs."

By late 2025, however, AI automation capabilities—embodied by coding agents—have rewritten that cost ledger:

Cross-platform translation and alignment: Agents can now reference an iOS implementation and automatically generate highly idiomatic Android code, and vice versa.

Specification and review automation: Shared spec documents, test cases, and adversarial review agents drastically reduce the maintenance cost of cross-platform feature parity.

Flattened learning curve: Engineers unfamiliar with a native platform can contribute effectively with agent assistance.

Native development still means two codebases, but the equation "two codebases = 2x headcount" no longer holds. When the writing and alignment cost of two native codebases is compressed by 70% or more, the "R&D headcount savings" that cross-platform frameworks provided can no longer offset the inherent risks they introduce: runtime complexity, Bridge/JSI glue layers, and non-native UI fidelity.

Re-deriving from first principles yields an unambiguous answer: return to Swift and Kotlin.

12 Weeks and 6 People: Greenfield Rebuild in the Agent Era

Traditionally, large-scale software rewrites avoid "greenfield rebuilds" because they often mean years of development stagnation and unlimited delivery risk. When Shopify originally migrated to React Native, it was forced into a painful "brownfield incremental replacement." Coding agents have now made "from-scratch rewrite" the preferred path.

1. Breaking the "Generating Garbage Code" Trap

Dumping a massive legacy codebase into an LLM to generate native code only yields unmaintainable waste. Shopify decomposed the migration into specialized agent workflows (Pi/Helix) :

Dedicated agents parse React Native source code and record runtime behavior.

Planning agents formulate Swift/Kotlin implementation roadmaps.

Every incremental checkpoint must pass visual comparison, unit tests, and adversarial code review agents before final human approval.

2. Reverse-Forced Architectural Innovation: Desktop Headless

Shopify discovered that AI coding speed is measured in seconds, but launching, rendering, and verifying UI in simulators takes minutes—verification became the new bottleneck. The team completely decoupled mobile business logic from UI, enabling it to run headless on desktop and expose a CLI to agents. The originally minutes-long verification loop collapsed to milliseconds, fully unleashing AI productivity.

Ultimately, 6 people in 12 weeks delivered the from-RN-to-full-native release. The performance gains are striking:

Cold start speed: improved 45%

Main-flow stutter/lag: reduced 62%

Memory footprint: reduced 38%

Shopify acknowledges some gains stem from product simplification, but the trend is unmistakable: native development's peak performance and seamless platform integration (Siri Shortcuts, widgets, Apple Watch support) have retaken the high ground.

History's Cycle: From HTML5's Lesson to AI-Era Selection Reconstruction

React Native's birth was itself a correction of "cross-platform technology compromise." In 2012, Mark Zuckerberg admitted "betting too heavily on HTML5 was Facebook's biggest strategic mistake." To solve Web's mobile jank and experience fracture, Facebook proposed "use JavaScript but render native controls"—the compromise that became React Native.

For a decade, engineering teams have walked the tightrope between "experience" and "cost." The React Native team spent six years rewriting the bottom layer to minimize the cost of that compromise. Yet just as React Native delivered a near-perfect scorecard in 2026, the rules changed:

AI did not defeat React Native; AI merely shattered the techno-economic assumption that "writing two codebases is too expensive."

Previously, technology selection was a fate line embedded in organizational bones. Migration costs of years and tens of millions meant that once React Native or Flutter was chosen, hiring, architecture, and deployment were deeply locked in. In the agent era, when a rewrite of thousands of pages costs weeks, technology selection ceases to be a one-time gamble and degrades into a dynamic, reversible engineering decision.

Conclusion

React Native will survive; Meta, Microsoft, Expo, and others continue heavy investment, and weekly downloads keep hitting new highs. For small-to-medium projects or validation MVPs, it remains an excellent choice.

But Shopify's exit sounds an alarm for the entire software engineering world:

Technological evolution is often brutal. Sometimes you did nothing wrong—you even spent six years perfecting yourself. But when the underlying economic ledger is rewritten by a new technology stack (AI agents), the foundation you survived on simply ceases to exist.