GitHub's engineering team used GitHub Copilot to rewrite a core part of Copilot, a project the team describes as unaffordable without agent assistance. The account is first-party and detailed. Agents handled large stretches of mechanical translation work, while engineers directed the shape of the effort and reviewed what came back. The ratio of agent work to human work shifted enough that the project became viable where it had not been before. For a UX leadership audience, the most useful signal here is about oversight at scale. A rewrite of this size generates a volume of changes no reviewer can hold in their head at once. The design challenge is giving engineers a clear view into what the agent did, and why, so they can make confident decisions about what to ship. Approval, review, and the ability to trace agent reasoning all become load-bearing at a scope that smaller agent tasks do not expose. The piece also illustrates a shift in what engineers do. They move from writing code to directing work, setting scope, and judging output. That shift changes what the interface to the agent needs to show. A status view, a list of changes, and a record of what the agent altered are no longer enough when the agent is producing hundreds of changes across a large project. Teams designing agent products for developers should expect this gap to surface fast.