Migrating the GitHub Copilot Runtime to Rust, with Help from Copilot

Migrating the GitHub Copilot Runtime to Rust, with Help from Copilot

By Stephen Toub, Distinguished Engineer at Microsoft

As AI-assisted development reaches new levels of sophistication in 2026, the story of how GitHub Copilot's own runtime was migrated to Rust—using Copilot itself—offers a compelling look at the future of software engineering: AI tools accelerating the very systems that power them.

Why Rust?

The original GitHub Copilot runtime was built on a mix of technologies, but as demand for Copilot grew, the limitations of the existing stack became apparent. Latency, memory safety, and concurrency bottlenecks pushed the team to consider a rewrite. Rust emerged as the ideal candidate:

• Memory safety without garbage collection
• Zero-cost abstractions
• Excellent concurrency support
• Strong ecosystem for performance-critical services

Using Copilot to Migrate Copilot

In a delightfully recursive twist, the team used GitHub Copilot to assist in the migration. Copilot helped:

• Translate idiomatic patterns from other languages into Rust
• Generate boilerplate for FFI bindings and async runtimes
• Suggest unsafe code alternatives with safer constructs
• Write unit tests to validate behavior parity

This wasn't blind automation—engineers reviewed and refined every suggestion—but it dramatically reduced the time spent on mechanical translation.

Challenges and Lessons Learned

1. Borrow checker friction: While Rust's borrow checker prevents entire classes of bugs, it required rethinking ownership models in the original code.
2. Async runtime choices: The team evaluated tokio vs async-std and ultimately standardized on tokio for its maturity and ecosystem.
3. Interop with existing services: Gradual migration via FFI allowed incremental adoption without a full rewrite.
4. AI hallucinations: Copilot occasionally suggested non-existent crates or outdated APIs, necessitating careful review.



Results



• ~40% reduction in memory usage
• 2x improvement in p99 latency
• Fewer runtime panics due to Rust's safety guarantees
• Faster onboarding for new engineers thanks to clearer code and tooling



The Bigger Picture




By 2026, AI-assisted migrations like this are becoming the norm rather than the exception. Tools like GitHub Copilot are not just writing code—they're reshaping how we approach large-scale system rewrites. The Copilot-on-Copilot migration is a proof point that the future of software is collaborative, recursive, and increasingly efficient.




---




Stephen Toub is a Distinguished Engineer at Microsoft, focusing on runtime performance and developer tooling.

via GitHub AI Blog