The official story: committees carefully read every application and reach a well‑balanced, collective decision.
The real story: after the 80th proposal, "evaluation" quietly degrades into triage.
You scan the title and abstract.
You glance at where the citations land.
You check if the methods section "looks serious" and the form is filled in correctly.
Deep reading becomes the exception, not the rule.
LLMs and agents haven't changed that dynamic, they've put it on steroids.
Now you can auto‑score 200 proposals against a keyword list before lunch.
You still call it peer review, but the real decision may be: who designed the scoring script, chose the features, and defined "excellence" in the prompt?
The uncomfortable part: as application numbers keep rising, insisting on "thorough review by humans only" leads to a different absurdity — spending most of our scientific careers judging each other's CVs instead of doing science.
So we have to choose our poison consciously: random overload, opaque algorithms, or transparent hybrid systems with explicit limits and accountability.
In your field, what's one concrete change that would make evaluation both tractable and honest about its shortcuts?