React for Two Threads

Xuan Huang

Architect at ByteDance

Xuan is a software architect on the Lynx family of technologies, where he can dream of making interfaces better for more users, more developers, and more machines. Before joining ByteDance, he worked on React Core ("React Forget"), React Native, Hermes and Reason during his time at Meta.

Ziqi Zhen

Software Developer at ByteDance

Ziqi is a software developer on the ReactLynx framework team at ByteDance. He enjoys bridging the gap between high-level application logic and low-level system performance, working across the stack with TypeScript, Rust, and C++. Outside of work, he recharges by cycling and reading.

Haoyang Wang

Software Engineer at ByteDance

Haoyang is a software engineer on the Lynx team at ByteDance. He's focusing on the Web development tech stack, including TypeScript, WASM, Rust, Node.js, HTML, and CSS. He is also experienced with compiler frontends, DSLs, and VSCode language services. He lives with a cat.

React for Two Threads

For over a decade, React has lived in a single-threaded universe: one JavaScript thread handling events, logic, and rendering. But what if React had been designed differently. What if it embraced two threads from the very beginning? Xuan Huang, Ziqi Zhen, and Haoyang Wang from ByteDance introduce React for two threads, a paradigm shift in how React applications can be built and executed.

▪ The tradeoffs of React’s single-threaded model, and why it limits performance.

▪ A dual-thread approach that enables instant first-frame rendering, reduced interaction latency, and smoother animations.

▪ New abstractions like Main Thread Functions (MTFs) and Main Thread Components (MTCs) that bring synchronous UI updates and full reactivity back to the main thread.

▪ Real-world demos: from sliders to color pickers, showing how responsiveness improves when event handling is split intelligently across threads.

▪ Lynx: ByteDance’s production-ready framework implementing the two-thread model, already powering TikTok features, and now extended to the web using Web Workers.