Generation Is Cheap, Review Is Expensive: How to Stop Shipping AI Slop — Gabriel Martinez, G2i

This post contributes to the understanding of software engineering practices in the age of AI by defining and addressing 'slop'—work that appears complete but lacks deep judgment or clear decision-making, often exacerbated by AI generation outpacing human review. It argues that code remains a liability, and AI agents can amplify existing codebase issues. The talk critiques incentives that push for faster merging over thorough review, the problem of large, unmanageable PRs, and the cognitive overhead of reviewing lengthy AI-generated documentation. It proposes solutions inspired by Rails conventions, emphasizing the use of diagrams as the primary review surface, the importance of small, reviewable changes, clear ownership, and the development of tools like ORC to manage these processes. The core message is that while AI generation is cheap, effective review and judgment are expensive and critical for shipping high-quality software.