GitHub Copilot app launched Stacked Sessions and Stacked PRs: infrastructure for multi-agent parallel development

When AI coding agents start to work in parallel like a team, what developers need is no longer "better single completion", but an infrastructure to manage "multiple parallel workflows". GitHub gave the answer on July 30: GitHub Copilot app introduced Stacked Sessions and Stacked Pull Requests (stacked PR), and Stacked PRs entered public preview on the same day.

What problems does stack workflow solve?

The traditional development model is "one change → one PR → merge and then do the next one". Efficient development of Agent-driven development requires parallelism: split a large change into multiple small PRs that depend on each other, and maintain multiple Agent sessions in parallel in the Copilot app - each session corresponds to a branch/change. This is the key to improving the efficiency of collaboration in large code bases: multiple Agent sessions work at the same time, each focusing on a change, and finally merged in dependency order.

From third party to official productization

Stacked PR was previously mainly supported by third-party tools, and GitHub will productize it into the Copilot app this time. Behind this is an important product judgment: with the popularity of AI coding Agents, developers are shifting from "single-threaded handwritten code" to "multi-Agent parallel development", and stack session/PR is the infrastructure to manage this new workflow. Whoever makes this set of capabilities an official first-class citizen first will be able to take advantage of the developer platform in the Agent era.

From an industry perspective, Stacked Sessions and Stacked PRs are important measures for GitHub to consolidate its status as a developer platform and compete against the Cursor/Devin ecosystem. The signal it revealed is that AI programming competition has extended from "model capabilities" to "workflow orchestration" - when one Agent is not enough and multiple Agents are needed in parallel, how to make their collaboration orderly and PR clear and reviewable is the real concern of enterprise-level teams. For domestic code hosting and development platforms, this prompts the next stage of competition: not whose AI can write code more efficiently, but who can provide complete engineering support for "multi-agent collaboration".

Several directions worth tracking in the future:

1. Public Preview to GA Rhythm: The official release time and stability of Stacked PRs.
2. Relationship with third-party stack tools: The impact on tools such as Graphite after official productization.
3. Conflict handling of multiple Agent sessions: dependency and conflict resolution mechanism between parallel changes.
4. Follow-up on domestic development platforms: Will platforms such as Gitee provide similar stack workflow capabilities.