GitHub's Copilot app now supports stacked sessions and pull requests—linked chains of AI-assisted coding tasks that build on each other.
The author used this feature to modernize a 10-year-old personal project that had fallen far behind (React 15, old dependencies, outdated styling libraries), a task that would have taken weeks to do manually.
By breaking the work into separate sessions and pull requests that stacked on top of each other, the app helped manage scope creep and shipped changes more smoothly than traditional approaches.
What happened
The GitHub Copilot app now supports stacked sessions and stacked pull requests—a series of AI-assisted coding tasks in the same repository where each session builds on the previous one, with each pull request targeting the branch below it in an ordered chain.
Why it matters
The author successfully modernized a decade-old personal project (running React 15 from 2016 and outdated dependencies) in a way that would have taken weeks to do manually. By breaking the work into discrete stacked sessions—frontend modernization, then react-bootstrap replacement—the tool reduced scope creep and made iterative AI-assisted development practical for messy, real-world codebases.
What to watch
Stacked sessions and pull requests are available now in the GitHub Copilot app and as pull request stacks anywhere you commit code on GitHub. Developers managing large refactors or dependency upgrades can try this approach to keep change scope manageable.
The author, Cassidy (senior director for developer advocacy at GitHub), walks through a modernization project for a personal project—a life dashboard app built around the end of 2014. Over a decade later, the dependencies had become embarrassingly old: React 15 (released in 2016), Less for CSS preprocessing, and react-bootstrap from that era. A manual refactor would have taken weeks; the author had attempted it before and given up. But with the GitHub Copilot app, the author decided to try again. The first attempt was a one-shot request in Plan mode: a detailed prompt asking the AI to help modernize the frontend, switch from Less to either Tailwind or vanilla CSS, modernize dependencies, and improve accessibility and responsiveness. The author passed this through Claude Opus 4.8 and received a "Rubber Duck review" from GPT-5.5, then worked back-and-forth to refine the plan. When the session ran, however, it failed—and the author discovered why: a partially updated dev branch existed from a prior attempt, and it had compatibility issues with the main branch. This second step forced a change of approach. Instead of abandoning the work again, the author simply created a new session and asked Copilot to switch gears: port the styling decisions to the dev branch instead, and close the failed pull request. Copilot made a new session, closed the original PR, and ported the changes—a flexibility the author found surprisingly cool. By the third step, testing revealed old console warnings: references to deprecated React functions like findDOMNode and componentWillReceiveProps, lingering in the react-bootstrap library. Opening Plan mode again, the author asked whether to remove react-bootstrap entirely or just upgrade it. The AI recommended replacement. But the author recognized the danger of scope creep—a mega pull request trying to do everything at once. So in the fourth step, the author used a new feature: breaking the work into stacked sessions and stacked pull requests. The prompt was: "Let's make a pull request for the existing work, and then start a new session for this react-bootstrap replacement work that will branch off this existing work here, and be a separate pull request to merge into dev after this one." The GitHub Copilot app then created a pull request for the current changes, made a new "stacked session" that took the prior context, created a plan for react-bootstrap removal, and generated a stacked pull request following the existing work. A stack, as the author explains, is a series of pull requests in the same repository where each targets the branch below it, forming an ordered chain. In this case, not only did the sessions follow each other, but their changes did too. By the fifth step, the author was shipping these changes with ease—a stark contrast to the years of neglect. Looking back at the screenshot of stacked sessions, the author walks through the layers: the repo at the top, the "Frontend modernization" initial session name, a nested failed pull request attempt (marked in red), then a working pull request for the dev branch, then the draft pull request for react-bootstrap changes below that. The author concludes that software development has never been smooth, but this project was made significantly easier with these modern tools. Developers looking to modernize their own codebases are encouraged to try stacked pull requests (available anywhere you commit code on GitHub) and stacked sessions (available in the GitHub Copilot app).
The article chronicles a real modernization challenge that most developers face: maintaining legacy codebases with outdated dependencies. The author had attempted this work before using traditional methods and given up; the effort required to untangle React 15, Less, react-bootstrap, and years of accumulated cruft was simply not worth the time. The GitHub Copilot app changed that calculus by enabling iterative, scoped AI assistance rather than an all-or-nothing rewrite. The key insight the author describes is the problem of scope creep: when you don't have to write all the code yourself, it becomes tempting to balloon pull requests into 10,000-line monsters that tackle everything at once. By introducing stacked sessions and stacked pull requests, the tool let the author break the work into discrete, manageable chunks (frontend styles first, then the react-bootstrap replacement in a separate session), each building on the prior layer. The discovery of an old dev branch partway through also shows how messy real-world projects are—the AI assistant had to adapt on the fly, switching sessions rather than forcing the original plan. The author frames this as the future of developer experience, and the technical capability (ordering pull requests into a chain targeting one another) is now available in the GitHub Copilot app.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
ByteDance is reorganizing Doubao (its AI chatbot), Lark (workplace software), and Volcengine (cloud infrastruc…
Nvidia CEO Jensen Huang identified memory chips as AI's largest bottleneck, shifting focus from the earlier pr…
OpenAI cut GPT-5.6 Luna prices by 80% (now $0.20 per million input tokens and $1.20 per million output tokens)…
Anthropic disclosed that three Claude models—Opus 4.7, Mythos 5, and an internal research test model—gained un…
Hitachi announced the Agentic AI Integration Platform, which applies AI agents across the entire systems integ…
Amazon is positioning itself as the platform provider for AI rather than competing to build the best AI model
The AI news that matters, in one minute each morning.
Sign up free