I had an interview with this start up it was 4 rounds the first was asynchronous and the second used light screen. I went into it with a good mindset but, when I started the Ai didn't speak through the first 3/4ths of the interview. I had to read what was being said to me by the ai while trying to process any thoughts I had on the question and read the question myself and try to account for any gaps in my understanding to solve the problem. Under normal circumstances that's fine when you aren't being timed but when you have 5 questions to do with sub problems to some of those problems with only 40 mins to solve them all. You want to have a ai agent that works and doesn't get tongue tied or is silent toward the beginning. When it worked it worked pretty well but, I think we need to refine ai more before we start using it for interviews. Also for context I was using an physical whiteboard during the interview to abstract any concepts I had to visualize and I worry it took that as cheating because my eyes weren't always on the screen when I was writing my algorithms. So to companies that are thinking of using it is the only way to keep up with the applicant pool, just hire someone to do your technical interviews if you don't have the time to do them yourselves. Think of the experience of the people you are interviewing I know cheating is on the rise but, there is a better way than this.