GitHub has moved the shared runtime behind a broad range of Copilot products from TypeScript and Node.js to Rust, using its own AI coding tools to carry out most of the port. The company says the resulting codebase exceeds 800,000 lines of production Rust and was integrated through 128 pull requests.
The runtime began as the engine for GitHub's cloud coding agent and later became the common agent harness used by the Copilot command-line interface, app and software development kit. GitHub says that same foundation now supports integrations across Visual Studio Code, Visual Studio, Copilot Code Review, Copilot Studio and several Microsoft productivity applications.
That expansion exposed costs in the original architecture. An application using the SDK had to start a separate Copilot CLI process, communicate through JSON-RPC and carry Node.js and the V8 JavaScript engine even when the host program was written in another language. GitHub estimated that arrangement imposed roughly 100 megabytes of minimum working memory for each client, as well as process-management and communication overhead.
Rust was selected because the runtime needed to embed through a C application binary interface, start quickly and use resources predictably. GitHub stressed that the decision was specific to those requirements, rather than a general argument that large TypeScript programs should be rewritten. Explicit management of lifetimes and shared state also introduced its own engineering complications.
The migration was shipped incrementally instead of being held for a single switchover. GitHub says AI agents wrote most of the new code, while a primary developer directed the work over several months and the wider team continued developing the product. Regressions were found and repaired during the staged rollout.
The scale grew far beyond the initial estimate. A planning exercise in May 2026 counted about 130,000 lines of TypeScript, but parallel product development and the work required to reproduce production behavior expanded the final Rust implementation. GitHub reports performance improvements measured in orders of magnitude, though its account does not provide one universal benchmark for every product using the runtime.
The project is a notable test of AI-assisted software maintenance because it involved a live, shared component rather than a standalone demonstration. Its significance rests not only on how much code agents generated, but on the controls around that output: small pull requests, continuous integration into the main branch and rapid correction of failures. The completed port also removes the requirement for every SDK consumer to supervise a second language runtime and process.
