GitHub rewrote its Copilot agent runtime in Rust with Copilot CLI

GitHub moved the agent runtime behind Copilot CLI, the Copilot app and the Copilot SDK from TypeScript on Node.js to Rust, with agents in the Copilot app and Copilot CLI writing most of the code. The port landed as 128 pull requests that replaced one component at a time while the rest of the team kept shipping, and GitHub reports it was done primarily by a single developer in a few months, work it says would have taken a whole team a year or two before agents.

The problem

The runtime was written in TypeScript on Node.js and V8, and the SDK reached it by spawning the CLI as a subprocess and talking JSON-RPC to it. Every SDK consumer in C#, Python, Go, Java and Rust therefore shipped a second language runtime, on the order of 100 MB of working set minimum, paid a process hop on every event and supervised two processes. The initial plan in early May 2026 estimated roughly 130,000 lines of TypeScript, but with code being pushed down from the terminal UI and hundreds of pull requests merging each week, about 430,000 lines of production TypeScript ended up passing through the port.

How they did it

1. Port in place rather than in a long-lived branch; each pull request replaces one component's TypeScript with a thin shim that calls the new Rust and deletes the old code in the same atomic change, so main stays shippable.
2. Start with pilot pull requests that set up the Rust workspace, toolchain, lint rules, CI, build pipeline and coding instructions, then port pure-logic helpers that already have strong tests, and work from the leaves inward, leaving the most coupled code for last.
3. Bridge the languages temporarily with napi-rs so TypeScript can call Rust and Rust can call back into TypeScript, and delete each shim as soon as its caller is ported.
4. Drive the work from the Copilot app and Copilot CLI; for large files a parent session splits the work and starts child sessions, each with its own branch and worktree, while subagents fan out exploration.
5. Run a review skill on every port that rebases, then has subagents on different models compare the old TypeScript with the new Rust line by line, and forbid agents from deleting or changing end-to-end tests.
6. Let agent merge (/pr auto in the CLI) answer review comments, fix CI failures and resolve conflicts, but have a human check how feedback was handled before merging.
7. Stop parallel sessions from overloading one machine by making a single session the gate that lets one session at a time build and test.

Results

• More than 800,000 lines of production Rust written, mostly by agents, across 128 pull requests that landed in main, as reported by GitHub.
• By August 21 the runtime was 100% production Rust, with 832,378 lines of production Rust and 468,689 lines of Rust unit tests, as reported by GitHub.
• Main shipped 135 releases during the porting window, 100 pre-release and 35 stable, so ported components reached users in small batches, as reported by GitHub.
• The runtime's performance improved by orders of magnitude, and the few regressions were found and fixed along the way, as reported by GitHub.
• The project was completed primarily by a single developer in a few months while the rest of the team kept expanding the runtime, as reported by GitHub.

As reported by the source (GitHub blog (engineering)); AgentGid did not measure these figures.

The agents used here

GitHub Copilot CLIGitHub (Microsoft) · Terminal coding agents

GitHub's standalone terminal coding agent (`copilot`, npm @github/copilot) that plans and edits code, runs commands and works with GitHub from the te…

Price

Free + $10/mo

Gid Score

49 #11 in Terminal

What it does, pricing, alternatives →

GC

GitHub CopilotGitHub (Microsoft) · AI IDEs and editor agents

AI coding assistant from GitHub offering completions, chat and agent mode in IDEs, a terminal CLI, a desktop app and a cloud agent that works on assi…

Price

Free + $10/mo

Gid Score

68 #2 in IDEs

What it does, pricing, alternatives →

Similar use cases

Showcase

How three Java Champions use GitHub Copilot for migration, testing and app building

Three experienced Java developers describe how they use GitHub Copilot in IDEs and agent workflows for feature work, a slide-deck migration, UI test reproduction and small apps. T…

As reported by the GitHub Blog (developer quotes): Vlad Mihalcea built an Optimizer event-history web app in two days, a task he expected to take wee…

GCGitHub CopilotContext7GitHub Copilot coding agentCode migration · Building apps

ShowcaseEmil Stenström

JustHTML: a Python HTML5 parser built with coding agents that passes html5lib tests

A developer spent his off-hours using Copilot agent mode with several models to build a dependency-free HTML5 parser, and wrote up every stage, including a full restart.

JustHTML is about 3,000 lines of Python with 8,500+ tests passing.

GCGitHub CopilotBuilding apps · Testing

Recipe

Generate unit and integration tests with GitHub Copilot

GitHub's tutorial uses Copilot Chat to write a unit test suite for a class, then integration tests that mock an external notification service, and then asks what coverage is still…

A unit test suite that covers edge cases, exceptions and data validation.

GCGitHub CopilotTesting