Slop is code you can't work with (dwyer.co.za)

A recent discussion has emerged around the term "slop," originally used to describe unrefined content but now increasingly applied to codebases, particularly those generated by AI models. The concept refers to code that is overly complex, poorly structured, and difficult to manage, often leading to unexpected bugs and inefficiencies. Although the definition of slop can vary from person to person, it generally describes situations where developers struggle to work with code, whether it’s AI-generated or human-written. This phenomenon highlights concerns over coding practices in the AI/ML community, especially as reliance on LLMs for rapid development becomes prevalent. The exploration of slop has significant implications for the software engineering process, particularly when using AI agents. Developers report navigating through a seemingly productive development phase that quickly devolves into confusion and breakdowns as they encounter unreliable code generation outputs. This unpredictability complicates task estimation and planning, making it hard to anticipate challenges. Interestingly, the author suggests that instead of viewing slop as an outright failure, it can serve as a flexible blueprint for further development. By using slop-as-a-spec, developers can learn from previous iterations and redesign projects without getting bogged down by initial errors, ultimately fostering a more iterative and experimental approach to coding in an AI-driven landscape.