This somehow assumes that the person above doesn't already do all the magic the people say work 100% of the time, and finds these issues. How does running into literally the same issues in a low-stakes project help?
replyI think it's a good assumption. Their comment implies they aren't harnessing the coding agent very strongly given the failure modes they mention.
replyDo you think it's better to experiment with high stakes projects? Do you think it'll be easier to avoid defaulting to line by line auditing of a project that is high stakes?