GitHub Copilot App Adds Stacked Sessions to Chain Multiple AI Coding Tasks Into Sequential Pull Requests

What's new

GitHub has introduced stacked sessions and pull requests in the GitHub Copilot app, a workflow feature that lets developers chain multiple Copilot coding sessions together so each one builds on the output of the last. Instead of asking Copilot to complete one large, monolithic task, developers can split the work into a sequence of smaller sessions, with each session generating its own pull request that depends on the branch created by the prior one.

GitHub detailed the feature in a blog post authored by a GitHub engineer who used stacked sessions to modernize a legacy personal codebase, according to the company. The post walks through breaking a large refactor into discrete steps — for example, upgrading dependencies, updating syntax, and restructuring code — with each step handled by its own Copilot session and surfaced as a separate, stacked PR.

How stacked sessions work

The stacking model mirrors the "stacked diffs" workflow already familiar to developers who use tools like Graphite or Sapling, where each PR builds on top of the previous one rather than all changes landing in a single branch. Applied to Copilot, this means:

• Each session in the stack starts from the branch produced by the session before it.
• Every session generates an independent pull request, keeping changes reviewable in smaller, ordered chunks.
• Developers can review and merge stacked PRs sequentially, or adjust an earlier session in the stack and propagate changes downstream.

GitHub did not disclose specific technical details on how conflicts between stacked sessions are resolved, nor did it specify which underlying model or models power the sessions. Pricing for the feature was not addressed in the post; it appears to be part of the existing GitHub Copilot app rather than a separately billed product.

Confirmed vs. claimed

The existence of the stacked sessions and pull requests feature in the Copilot app is confirmed by GitHub's own blog post. The specific example — modernizing an old codebase — is presented by GitHub as a real-world demonstration, but performance claims (time saved, code quality improvements) are not quantified in the source material and should be treated as anecdotal rather than benchmarked.

What this means

Stacked sessions address a real limitation of single-shot AI coding agents: large, multi-step refactors are hard to review and hard to safely automate in one pass. By breaking work into a dependency chain of sessions and PRs, GitHub is borrowing a code-review pattern already popular among engineering teams and applying it to agentic coding workflows. This makes Copilot-generated changes easier to audit incrementally rather than reviewing one sprawling diff.

The move also signals competitive pressure from tools like Cursor (Anysphere), Cline, and Sourcegraph's Amp, which have been pushing agentic, multi-step coding workflows. Whether stacked sessions meaningfully reduce review burden or introduce new complexity — such as managing rebases when an early session in the stack changes — will depend on real-world adoption. GitHub has not published data on error rates, merge conflict frequency, or time savings, so early users should treat the feature as workflow tooling rather than a fully validated productivity gain.