As others have said, the premise of the test is wrong. Of course if you use tools for tasks they're not suited for, the results will be poor.

If you use AI to help build small well-described subsets of the problem, which you fully understand, you can considerably reduce development time.

If you ask for too much, and get nonworking garbage, you shouldn't spend hours debugging it--you throw that away, and design a solution yourself, then ask the AI to do smaller pieces. Keep doing it until you get into the sweet spot where AI does it right (or you realize that it's better to do it yourself). Shifting your work to more software design (which you always had to do) and less line-by-line coding is where AI provides its benefits.