In my mind this hinges on whether LLMs become capable of actual comprehension of the code they’re writing. If not, for serious projects (especially those which are mission-critical), it still makes sense to optimize for human understanding when selecting languages so the code can reasonably be verified.