Someone should try this with the “Ralph Wiggum loop” approach. I suspect it would fail spectacularly, but it would be fascinating to watch.
Personally, I can’t get meaningful results unless I use the tool in a true pair-programming mode—watching it reason, plan, and execute step by step. The ability to clearly articulate exactly what you want, and how you want it done, is becoming a rare skill.
Given the quality of their existing test suite I'm confident the Ralph Wiggum loop would produce a working implementation... but the code quality wouldn't be anywhere near what they got from two weeks of hands-on expert prompting.
Sure yeah, I can buy that, but that would be like collecting tech debt for generations.
https://ghuntley.com/loop/ and https://github.com/anthropics/claude-code/blob/main/plugins/...
Looks like it's been renamed to ralph-loop for legal reasons. :D
https://github.com/anthropics/claude-plugins-official/pull/1...