Eight prompts, ten minutes of agent time, one failed prediction, and two bugs introduced by the human-in-the-loop.
I started writing this article to better understand Loop Engineering which is the new Prompt Engineering (how 2025 that is now). I thought this would mostly involve understanding how to create an effective coding harness, but for completeness I tried one-shotting the project with a single prompt. In some ways the one-shot version was better than the loop engineered game, but in some ways it is worse. The more complex the project the more effort you need to put into the harness, but if models keep improving (and are allowed to be exported)…
Leave a Reply