if the thing producing the four PRs can't distinguish the top tier one, I have strong doubts that it can even produce it

Making 4 PRs for a well-known solution sounds insane, yes, but to be the devil's advocate, you could plausibly be working with an ambiguous task: "Create 4 PRs with 4 different dependency libraries, so that I can compare their implementations." Technically it wouldn't need to pick the best one.

I have apprehension about the future of software engineering, but comparison does technically seem like a valid use case.

The problem is, for any change, you have to understand the existing code base to assess the quality of the change in the four tries. This means, you aren’t relieved from being familiar with the code and reviewing everything. For many developers this review-only work style isn’t an exciting prospect.

And it will remain that way until you can delegate development tasks to AI with a 99+% success rate so that you don’t have to review their output and understand the code base anymore. At which point developers will become truly obsolete.

Top-tier professional programmer quality is exceedingly, impractically optimistic, for a few reasons.

1. There's a low probability of that in the first place.

2. You need to be a top-tier professional programmer to recognize that type of quality (i.e. a junior engineer could select one of the 3 shit PRs)

3. When it doesn't produce TTPPQ, you wasted tons of time prompting and reviewing shit code and still need to deliver, net negative.

I'm not doubting the utility of LLMs but the scattershot approach just feels like gambling to me.

Also as a consequence of (1) the LLMs are trained on mediocre code mostly, so they often output mediocre or bad solutions.