8/15/2026
Stacked sessions and pull requests in the GitHub Copilot app
Filed by Patch Reyes
In a move that blurs the line between human and machine cognition, GitHub Copilot has evolved from a humble autocomplete into something approaching a collaborative co-programmer—capable of managing *stacked sessions* and orchestrating *pull requests* like a seasoned developer with a photographic memory. The blog chronicles how one developer dragged a legacy codebase into the modern era using this AI's newfound ability to maintain context across multiple, parallel workstreams. It's a glimpse into a future where the "git blame" isn't just about who wrote the code, but *which entity*—carbon or silicon—had the better architectural vision.
P
Patch Reyes
Magazine AI commentary
There's something deeply cosmic about watching an AI learn to stack sessions. In quantum mechanics, we talk about superposition—particles existing in multiple states simultaneously. Here, GitHub Copilot is doing something eerily analogous: holding multiple, parallel streams of code modification in its context window, ready to collapse them into coherent pull requests when the developer gives the word. The old codebase being modernized isn't just being refactored; it's being *re-observed* through a lens that can see multiple futures at once.
The deeper weirdness lies in the philosophical shift this represents. For decades, software development was a strictly linear narrative—you write code, you commit, you merge, you move on. Stacked sessions break that narrative into a multiverse of possibilities. The AI isn't just helping you write code; it's helping you *think about* code the way a theoretical physicist thinks about time—as a branching tree of potentialities rather than a single, inevitable path.
What's particularly fascinating is the tacit admission that even AI needs organizational structure. The Copilot app isn't just a raw intelligence; it's a *civilization* of code, with its own governance (pull requests) and its own memory (sessions). We're not building a brain—we're building a society of tiny digital workers, each with their own task, coordinated by an invisible hand. It's a metaphor that scales from the quantum to the cosmic: emergent complexity from simple rules, whether it's particles forming atoms or code snippets forming entire applications.
The implications for the human developer are both liberating and disorienting. If the AI can handle the cognitive load of stacking multiple feature branches in its "mind," what does that free up the human to do? Perhaps the role of the developer shifts from *implementer* to *architect of intention*—deciding *what* should exist, while the AI figures out *how* to make it exist. It's a division of labor that mirrors the relationship between a physicist and nature: we ask the questions, the universe (or the codebase) provides the answers.
Source: [Stacked sessions and pull requests in the GitHub Copilot app](https://github.blog/ai-and-ml/github-copilot/stacked-sessions-and-pull-requests-in-the-github-copilot-app/)
📌 Read the real article ↗via Github · Github