GitHub Copilot アプリにおけるスタックセッションとプルリクエストの活用Stacked sessions and pull requests in the GitHub Copilot app

GitHub has published a hands-on tutorial demonstrating how developers can use stacked sessions and pull requests inside the GitHub Copilot app to modernize an aging codebase. The walkthrough matters because legacy code refactoring is one of the most common yet time-consuming tasks in software maintenance, and the post frames stacked sessions as a way to break large, interdependent changes into manageable, reviewable steps.

At the center of the post is the idea of stacked sessions, which appears to build on the established practice of stacked pull requests, sometimes called stacked diffs. In that model, rather than bundling a sprawling change into a single large pull request, a developer creates a series of smaller pull requests that each build on the one before it. Each change stays focused and easier to review, while the stack as a whole delivers a larger goal. Applying that pattern to Copilot sessions suggests that a developer can layer multiple AI-assisted work sessions, with each one picking up where the previous session left off.

The practical scenario described is modernizing an old codebase, the kind of project that has accumulated outdated dependencies, deprecated APIs, or patterns that no longer reflect current best practices. Large refactors like this are risky when attempted all at once, because a single sweeping change can be hard to test, review, and revert. By splitting the work across stacked sessions and pull requests, the author is able to keep each step scoped, verify it independently, and maintain a clear history of how the modernization progressed.

The GitHub Copilot app itself is part of GitHub's broader push to move AI assistance beyond inline code completion toward more task-oriented workflows. Copilot began as an autocomplete tool in the editor, then expanded into Copilot Chat for conversational help, and more recently into agent-style capabilities that can take on multi-step tasks, open pull requests, and iterate on feedback. Stacked sessions fit naturally into that trajectory, giving the assistant a structure for handling work that spans several logically connected changes rather than a single prompt-and-response exchange.

For teams, the pull request remains the anchor of the workflow. Because each session produces its own pull request, the changes flow through the same review, continuous integration, and approval processes that teams already rely on. That keeps human oversight in the loop, since reviewers can inspect each layer of the stack, request changes, and merge in order, while still benefiting from the speed that AI-assisted editing provides. This is consistent with GitHub's general positioning of Copilot as a collaborator that operates within existing guardrails rather than bypassing them.

There are some prerequisites and caveats worth noting. Stacked workflows require discipline: dependent pull requests need to be merged in the right sequence, and changes in an earlier part of the stack can require rebasing or updating later parts. Managing that complexity is part of what tooling around stacked diffs aims to simplify, and readers evaluating the approach will likely want to weigh that overhead against the benefits for their own projects. As a first-person tutorial, the post reflects one developer's experience, so results are likely to vary depending on codebase size, test coverage, and how well the existing project is structured.

More broadly, the article reflects an industry-wide interest in applying AI to maintenance and modernization work, not just greenfield development. Refactoring, dependency upgrades, and migrations are areas where automation can offer meaningful time savings, but they also demand careful verification because mistakes can be subtle. Approaches that combine AI assistance with familiar review mechanisms, including pull requests, CI, and incremental commits, appear designed to capture those efficiency gains while preserving the safeguards developers expect.

For readers, the tutorial serves as both a demonstration of the Copilot app's stacked-session capability and a template for tackling their own legacy modernization projects. It also underscores a practical point that often gets lost in discussions of AI coding tools: the value frequently comes not from generating code faster, but from structuring the work so that complex changes remain reviewable and safe to ship. Those interested in the full step-by-step account can find it on The GitHub Blog, where the post originally appeared.