GitHub ships Stacked pull requests to tame AI-sized code changes

The case for stacking small pull requests is not new — external tools such as Graphite and Sapling have offered stacked workflows for years. What changed is that AI coding agents, which run long loops of observing, planning, executing and evaluating, now produce large volumes of code changes at once, and a single giant diff forces reviewers to hold too much context in their heads. GitHub folding stacked PRs into the platform as a standard feature reads as an industry answer to that mismatch between how much code AI generates and how much a human can review at one time.

Google Cloud's DORA research program's 2025 "State of AI-assisted Software Development" report describes AI as an amplifier that magnifies an organization's strengths and weaknesses alike. DORA found AI adoption helps delivery throughput and product performance but remains linked to delivery instability, and interprets this as downstream weaknesses surfacing as AI speeds development up. Absent strong automated tests, mature version control and fast feedback loops, rising change volume tends to destabilize delivery.

Developer experience is a precondition for getting value from AI rather than a perk: fast CI, written-down implicit knowledge, and small review units are long-standing software engineering practices that now double as the conditions AI agents need to work well. GitHub's stacked-PR feature is one concrete piece of that groundwork.