# Shopify goes back to Swift and Kotlin > Shopify bet big on React Native in 2020. Coding agents broke the assumption behind that bet, so the company is rebuilding native. - Title: Shopify goes back to Swift and Kotlin · Shopify | Shortify - Summary: Shopify bet big on React Native in 2020. Coding agents broke the assumption behind that bet, so the company is rebuilding native. When AI changes the cost of… - Keywords: mobile, react native, ai agents, shopify, engineering, technology, startups, goes, back, Swift, Kotlin - Source: Shopify — https://shopify.engineering/back-to-native - Published: 2026-09-10T00:00:00+00:00 - Read time: 3 min - Topics: mobile, react native, ai agents, shopify, engineering, technology, startups ## Shopify is going native After six years on React Native, Shopify is rebuilding its mobile apps in Swift and Kotlin. Coding agents changed the math twice. > Native is now the future of mobile at Shopify ## The 2020 bet worked React Native let Shopify build features once, brought in developers with no mobile background, and ended the constant chase for feature parity. > that bet has been extremely successful ## Last year's answer was yes In January 2025 the company publicly recommitted to React Native. That was honest given what was known then, and the framework still works well. > the future of React Native was bright ## One assumption collapsed Building the same feature twice used to mean twice the work. Better coding models made that stop being true for Shopify's apps and team. > LLMs changed one of the core assumptions behind our 2020 decision ## Success is not a lock-in Shopify reevaluated the whole mobile stack from first principles rather than defend a choice because it had paid off before. > We don't hold on to a decision just because it was successful at the time. ## The tax they accepted React Native cost real time in performance tuning, foundational fixes, and chasing framework and dependency updates. The benefits outweighed it. ## LLMs since 2021 Shopify started using models for features, bug fixes, and code review a year before ChatGPT. The work they trusted models with grew as the models did. > one year before ChatGPT! ## The prototype surprise Rebuilding core parts of the biggest apps in Swift and Kotlin with agents worked better than expected, including porting an iOS feature to Android by reference. ## Parity got cheap Shared specs, tests, and review checkpoints cut the cost of keeping two platforms in sync, and agents helped developers contribute outside their main stack. ## Two platforms, still The duplication cost has not vanished. Agents now absorb enough of the implementation, translation, testing, and review that it stopped being decisive. > that cost has not disappeared ## Not a speed complaint Shopify says its React Native apps are fast. The pull is fewer layers between code and platform, closer access to first-party tooling. > React Native apps can be fast. Ours are. ## Skia gets a new home Shopify sponsors React Native Skia through end of 2026. William Candillon will fork it under a new name, and the original repo gets archived. ## FlashList needs a steward At roughly 2M downloads a week, FlashList keeps getting critical compatibility fixes while Shopify talks to companies about long-term ownership. > ~2M downloads/week ## Restyle winds down Smaller user base, so the repo is archived. It keeps working through 2026, then maintenance stops. Forks welcome, with handover help offered. ## Greenfield, not gradual Last time Shopify migrated in place to avoid pausing feature work. This time agents made full rebuilds fast enough that a clean slate won. ## Shop rebuilt in 12 weeks The Shop app went from proof of concept to a fully native app in the stores in twelve weeks. The Shopify app, 300+ screens plus widgets and a Watch app, ships later this year. ## One-shotting produces slop Pointing an LLM at the React Native codebase, even with frozen specs and task files, yields piles of unmaintainable code that cannot ship. > it doesn't work ## Helix gates every step Helix slices a screen into small checkpoints. Each must pass tests, a visual match, two adversarial reviewers, and a human before the next begins. > an imperfect attempt simply cannot move forward until it becomes a good result ## Simulators were the bottleneck Agents change code in seconds but waited minutes to test it, stuck driving simulators through accessibility trees and screenshots. > It doesn't matter how good the model is if it can't test its work quickly ## Architecture for agents Business logic is decoupled from UI and runs headless on desktop behind a CLI, so agents inspect state, navigate, and act in milliseconds and work for hours alone. ## Key takeaway When AI changes the cost of an engineering tradeoff, revisit the decision that tradeoff justified, even a successful one.