AI, Slop, and Where I Draw the Line in Retro Programming

Summary

The author argues that the important distinction in AI-assisted programming is not whether an AI tool appears anywhere in the workflow, but how much human effort, direction, responsibility, and review remain. They use AI tools including Cursor and Claude for tasks such as debugging, Docker setup, GitHub automation, backup scheduling, transcription, information retrieval, and filtering the Worth a Watch feed, while still selecting and editing every published story. The author is concerned about careless low-effort output, loss of human judgment, environmental and social harm attributed to AI companies, and the unlicensed use of protected intellectual property in AI training. Their working definition of programming includes giving detailed instructions, steering toward a goal, and checking the result; a vague request to generate a game does not meet that standard. Because fully AI-made games, ports, decompositions, and FPGA cores are judged not to contain sufficient human creative or engineering work, Worth a Watch currently excludes them. The author notes that hackathons, game jams, and open-source projects are also dealing with large volumes of generated submissions and code, while acknowledging that the policy may change as practices evolve. The article’s broader recommendation is to ask AI to explain work rather than simply perform it, so users retain enough understanding to detect errors.