AI coding agents can generate features fast, but can package them in one giant,

GitHub's proposed remedy for an oversized AI-generated code change is to divide it along dependencies. Its worked example puts the data foundation at the bottom of a stack of pull requests, then layers the API and interface work above it. Reviewers examine the foundation first, so each later review builds on code they already understand. When a lower layer changes, the dependent branches must be updated and their checks rerun. The benefit is a smaller, coherent review target at each step, while preserving the relationships between changes.