AINEWS
The port killed a ~100MB Node.js runtime that every Copilot install used to ship and exposed the new Rust engine as a C library that six SDK languages can embed directly.
Microsoft's Copilot used to ship a second programming language inside every install: a bundled copy of Node.js, weighing roughly 100 megabytes per client. The company's fix, disclosed this week, was to rewrite that engine in Rust and expose it as a C library. The reported cost: about $120,000, roughly a few months of a junior engineer's loaded salary. The reported crew: one developer, working with an AI agent.
The disclosure came from GitHub's engineering blog on September 16, which detailed a port of the Copilot runtime from TypeScript on Node.js/V8 to more than 800,000 lines of production Rust. The work landed across 128 pull requests over a few months, mostly from a single developer. The post's author notes, with some restraint, that the Rust compiler remains unperturbed by the change.
The architectural driver was size. Every Copilot SDK consumer, and there are now six (C#, TypeScript, Python, Rust, Go, and Java), used to have to either bundle Node.js or rely on a system install, paying a working-set penalty of roughly 100 MB per client. A new C ABI lets all six front-ends embed the engine in-process, deleting the second-language runtime tax. The same runtime backs the Copilot CLI, the Copilot app, VS Code, Visual Studio, the cloud agent, Copilot Code Review, Copilot Cowork, Copilot Studio, and Office (Excel, Outlook, PowerPoint, Word).
The agentic workflow, an AI that plans, edits, and verifies code across many steps with humans setting the goals, was the leverage, not the labor. The single developer used both the GitHub Copilot app and the Copilot CLI as the agentic authoring tools. The GitHub post is explicit that verification, review, and architectural decisions remained human work. The numbers in the post are net-flow estimates from the author: roughly 430,000 lines of TypeScript actually passed through the port, while about 1.2 million Rust lines entered and roughly 365,000 left.
The $120,000 figure comes from The Register's coverage on September 18, and is not visible in the GitHub post itself. Treated as a Register-sourced claim, the cost still lands in a defensible range: a few months of loaded engineering time, the kind of work that historically took a runtime-team port over a release cycle. The Register's headline frames the project as a budget test of agentic coding. The GitHub post frames it as a tax-deletion exercise. Both reads fit the disclosed facts.
The performance claim is the part to handle carefully. GitHub says the new runtime improves performance "by orders of magnitude." That is the company's own characterization, not a third-party benchmark, and it covers startup latency, memory footprint, and tail latency in roughly that order of importance. Specific multipliers (10x, 100x) are not published, and the absence is the news, not a flaw in the report.
A community-maintained Rust SDK at github.com/copilot-community-sdk/copilot-sdk-rust is in technical preview this week, talking to the runtime over JSON-RPC on stdio or TCP and requiring Rust 1.85 Edition 2024. The C ABI exposure is the more interesting move: it turns a previously Node-anchored service into something six different language runtimes can hold as a library.
The unit-of-account question is the one the industry is going to keep arguing about. A 128-PR, 800,000-line port in a few months by one developer, with an agentic loop, is a different shape of work than the same port without one. GitHub is not claiming the AI did the work alone. The post is explicit that humans set the architecture, reviewed the diffs, and ran the verification. What GitHub is claiming is that the second-language runtime tax is gone, and that six language SDKs are now first-class consumers of a Rust engine. The tax was the story. The cost was the receipt.