The Copilot agent runtime underlying GitHub Copilot CLI, GitHub Copilot app, and GitHub Copilot SDK was initially developed for the GitHub Copilot cloud agent (CCA) using TypeScript, Node.js, and V8. With the help of GitHub Copilot app and Copilot CLI, GitHub rewrote the runtime from scratch in more than 800,000 lines of production Rust code. AI agents generated most of the code, with the changes gradually merged into the main branch through 128 pull requests. Performance improved by several orders of magnitude, while the resulting regressions were quickly resolved. The project, which might previously have taken a team one or two years, was completed in a few months, largely by a single developer, while other team members continued expanding the system’s capabilities.
The shared runtime supports GitHub Copilot CLI and GitHub Copilot app, as well as the latest versions of VS Code, Visual Studio, CCA, Copilot Code Review (CCR), Copilot Cowork, Copilot Studio, Excel, Outlook, PowerPoint, and Word. Many of these products initially developed their own agent loops but later switched to GitHub Copilot SDK, which provides access to the Copilot agent runtime. The aim was to enable security, reliability, performance, and intelligence improvements to be shared across products.
However, although the architecture based on TypeScript, Node.js, V8, Ink, and React was sufficient for a console application, it created limitations in environments requiring fast startup, low memory usage, and high server density. In addition, in the rapidly developed initial CLI version, the terminal interface and runtime were intertwined rather than structured as separate layers.
Why it matters
The impact of this change extends beyond a single developer tool; the broad group of products using the shared runtime can benefit from improvements in security, reliability, performance, and intelligence through the same infrastructure. Addressing the limitations of the previous technology stack in environments requiring fast startup, low memory usage, and high server density is relevant to teams bringing the runtime to different products. The fact that the process was carried out over several months, largely by a single developer and AI agents, shows that the division of labor and review practices can change in large-scale production code transformations. The integration of the changes through 128 separate pull requests and the resolution of regressions demonstrate that speed does not eliminate the need for incremental review. However, the results are based on GitHub’s own technical account; no information is provided about the long-term maintenance burden of the new architecture.
Background
GitHub is not a new name in the FikirPilot archive: over the past 90 days, we published 3 articles mentioning the name, the latest dated September 11, 2026.
Term: agent
An AI agent is software that calls tools and carries out multi-step tasks to achieve a goal rather than producing a single response.