GitHub Rewrites Copilot Core in Rust with Copilot

GitHub announced that the agent engine powering Copilot CLI, the Copilot application, and the Copilot SDK was rewritten from TypeScript/Node.js/V8 to Rust—more than 800,000 lines of production code.

Copilot performed the rewrite itself: agents wrote most of the code.

At the same time, a batch of ecosystem changes was released: automatic closure of code review comments, enterprise-level allowlists for MCP servers, access to Claude Opus 5 in Copilot, and the transition of GitHub MCP Server to a new protocol.

Each of these events is routine on its own.

Together, they show that AI tools in development have moved from the demo phase to the infrastructure phase, where they face production-level requirements.

Companies have used LLMs for years to autocomplete lines and create small patches, while keeping the core of their systems—anything that cannot be broken without harming thousands of users—outside the agents’ reach.

The logic is clear: an agent that confuses types or loses context during a long task is unacceptable in a runtime serving millions of requests. GitHub removed this limitation in its own most critical component.

The agent runtime is what every call from Copilot CLI, the application, and the SDK passes through. Rust was chosen for a reason: the language catches memory-ownership errors and data races at compile time, requiring a level of code discipline that is difficult to fake.

Rewriting 800,000+ lines from TypeScript to Rust while keeping the runtime operational was a task that, just two years ago, would have been assigned only to a senior team for several quarters.