Large AI-generated pull requests are hard to review, but GitHub's stacked pull requests feature offers a solution. By decomposing a feature into a layered dependency chain of small, focused PRs, each scoped to a single concern (data, API, wiring, UI), teams can assign specialized reviewers per layer and maintain cleaner code review workflows. The post walks through a concrete example of adding product search to a shopping assistant, showing how to use the `gh stack` CLI extension to initialize, push, and rebase a four-layer stack. It also covers how coding agents can be taught to create and manage stacks autonomously, and explains the reviewer's directional workflow: read top-down for context, review bottom-up for correctness. Key operational details include how to safely rebase a diverged stack using `gh stack rebase` instead of the web UI button to preserve signed commits.
