Having thousands of applicants is only an issue if you give yourself the contrived problem of hiring the best person who sent you a resume. Your true objective is to strike a balance between cost of search and hiring someone from the top N% of potential people. Nobody has ever walked into a grocery store and bemoaned that there's no way they could locate the ripest banana in the building. You pick a number, evaluate that many at random, move on.

I think it galls people that they are likely cutting the best candidate out of the sample, but to be real: you don't have a magic incredibly sensitive, deterministic and bias free hiring method that can reliably pick the single best candidate out of thousands anyways. Any kind of cheapo ai-driven interview step you run is very possibly doing worse things to your sample than just cutting it down to size.

One of the refreshing things about the Amazon/AWS hiring approach was basically this. Did we agree this person can do the job? First one to get to a yes gets an offer. No interviewing all the candidates and stack ranking and trying juggle them to have a plan A and plan B (though people could influence that somewhat through scheduling). First qualified candidate succeeds and everyone gets back to work.