The company that championed React Native as a way to write mobile code once across platforms is now reversing course. Shopify announced Thursday that it is moving toward fully native development for its mobile applications, with AI agents handling much of the heavy lifting. The transformation began with Shop, its consumer-facing app, which progressed from proof of concept to production-ready native code in just 12 weeks. The merchant application—a sprawling 300-plus screen platform with extensive iOS integration—represents the next major undertaking.

The 2020 decision to adopt React Native was sound at the time and served the company well for years. Shopify does not view that choice as a failure; rather, it fulfilled its intended purpose. The shift away from cross-platform frameworks was not telegraphed in advance; as recently as January 2025, company leadership continued discussing the platform's role in future development.

Agents replaced cross-platform tradeoffs

The calculus changed as AI capabilities matured. By late 2025, Shopify discovered that agents could examine how a feature functioned on iOS and construct the corresponding Android implementation, or vice versa. This capability allowed engineers to contribute to platforms outside their expertise, with agents managing platform-specific complexities.

LLMs changed one of the core assumptions behind our 2020 decision

Mustafa Ali, Shopify's director of engineering

The Rewrite Economics Keep Changing

Shopify is far from alone in deploying agents for substantial code transformation tasks. Jarred Sumner, creator of Bun and a member of Anthropic's technical team, orchestrated a migration of the JavaScript runtime from Zig to Rust using 64 parallel Claude Fable 5 instances. The undertaking involved roughly one million lines of code and took 11 days, with an estimated API cost of $165,000. The existing test suite passed across all six supported platforms following the merge.

A year ago, that would have taken a small team multiple quarters but today it's an 11-day sprint supervised by one person.

These productivity improvements are not distributed uniformly across organizations. As reported recently, AI agents can generate additional work elsewhere within engineering teams even while accelerating specific development activities. However, they are fundamentally reshaping the economics of decisions that were previously costly and difficult to reverse.

In May, developer Simon Willison documented a conversation with an engineer whose organization had used coding agents to consolidate separate iPhone and Android applications into a single React Native codebase. When Willison questioned the logic of consolidation given that agents were making independent codebases easier to maintain, the engineer expressed no concern about vendor lock-in. The framework met their needs, and if circumstances shifted, they could transition back to native development. Shopify is executing a comparable strategy, though with substantially greater complexity and higher stakes.

Zero-shot prompting produced junk

Shopify's initial approach was straightforward: feed the existing React Native codebase to an LLM and request a native rewrite. The outcome was disappointing. Ali characterizes the generated code as slop, and introducing additional intermediate steps failed to resolve the fundamental issue.

Even when engineers directed the model to generate specifications and task files before writing code, the output remained unmaintainable. The volume of generated code exceeded what teams would realistically support.

Helix breaks migrations into pieces

Shopify developed Helix, an internal framework for orchestrating agent-driven rewrites. Rather than attempting wholesale conversion, Helix processes migrations screen by screen, subdividing each screen into manageable components. The system continuously validates output against the existing application while separate agents identify potential issues before human approval. Feedback from reviews is retained and informs subsequent work.

This methodical approach echoes the incident when Shopify's CEO publicly objected to Claude Code after engineers deployed agent-generated code with insufficient review. Helix embodies comparable caution by requiring agent output to demonstrate reliability before deployment.

Codebases built for agents

The migration effort revealed a practical constraint: agents can modify code in seconds, but validating changes in a mobile simulator requires substantially longer—often twice as much time. Shopify addressed this by decoupling business logic from user interface components and enabling agents to interact with the application through a desktop CLI. Validation checks that previously consumed minutes now execute in milliseconds, while the same interface can drive a simulator when necessary.

Some checks that once took minutes now happen in milliseconds, while the same interface can control a simulator when needed.

Most existing codebases were not designed with AI agents in mind. Shopify is constructing its architecture around agent capabilities and plans to evaluate the native applications partly based on how much work agents can eventually perform independently.

Source: The New Stack