It is easy to agree with basically everything in this article.
But IMHO it lacks an important aspect: Isn't it possible to leverage LLMs to actually improve peer review? E.g., by authors using LLMs before review to improve their manuscripts in ways historically done by human feedback. Authors have always critically evaluated such feedback.
Lots of tasks that reviewers used to perform can be done before review and thus allow reviewers to focus more on their core tasks?
https://www.timeshighereducation.com/opinion/ai-not-peer-so-it-cant-do-peer-review