Shopify Engineering·September 21, 2026Shopify's Helix: A System for AI-Assisted Mobile App Migrations
Shopify's Helix is an internal system designed to facilitate large-scale mobile application migrations from older frameworks to Swift and Kotlin using Large Language Models (LLMs). This system focuses on breaking down complex migrations into manageable, shippable chunks, incorporating continuous integration and strict quality gates to ensure code stability and developer productivity during AI-driven code generation.
Read original on Shopify EngineeringThe Challenge of Large-Scale Code Migrations
Migrating a mature, large-scale mobile application presents significant challenges, especially when moving between different programming languages or frameworks. Traditional methods often lead to large, risky, and difficult-to-merge codebases. Shopify faced this with its mobile app, aiming to transition from legacy code to modern Swift and Kotlin, and leveraged LLMs to accelerate this process.
Helix: An Orchestration System for LLM-Powered Rewrites
Helix acts as an orchestration layer for managing the lifecycle of LLM-generated code. Instead of generating the entire application at once, Helix breaks the migration into 'checkpoints' – small, independent units of work. Each checkpoint represents a functional slice of the application, allowing for incremental migration and continuous integration. This approach reduces the risk associated with AI-driven code generation by enabling frequent testing and validation.
Key System Design Principle: Incremental Change
Breaking down a large, complex task (like a full application rewrite) into smaller, manageable, and independently verifiable units is a core principle in system design. This minimizes blast radius, improves debuggability, and enables continuous delivery, even when using novel techniques like AI code generation.
Quality Gates and Feedback Loops
A critical aspect of Helix is its emphasis on strict quality gates. Each LLM-generated checkpoint undergoes automated testing, code linting, and potentially human review before being integrated. This feedback loop is crucial for guiding the LLMs and ensuring the generated code meets production standards. The system allows for rapid iteration, where LLMs can learn from failed checkpoints and refine their generation process.
- Continuous Integration: Automates testing and validation for each small code change.
- Automated Quality Checks: Static analysis, unit tests, and integration tests applied to LLM-generated code.
- Human-in-the-Loop: Optional or mandatory human review for critical components to ensure correctness and maintainability.
- Rollback Mechanisms: Ability to easily discard or revert problematic AI-generated code segments without impacting the entire migration.