AINEWS
Asynchronous AI interviews are sold as access for time poor workers, but they push the clock onto the candidates with the least margin.
At 9:30 pm on a Tuesday, Tim Millard sat down at his Atlanta kitchen table, opened an email from a media-relations role, and pressed record.
The prompt on his laptop gave him thirty seconds to think, two minutes to answer, and permission to re-record at will. There was no human on the other end. The slot was open-ended. He could have done it at 11 am or at 1 am, or pushed it off entirely. He picked 9:30 because that was the only window the day had. In March, after recording his take, he received an automated rejection by email.
He's not alone, and the data suggests the kitchen-table slot is now a measurable slice of the hiring funnel rather than a curiosity. Ribbon, a voice-AI recruitment platform used by more than 500 companies, reports that 24 percent of its AI interviews now run between 10 pm and 2 am local time. For Ribbon's manufacturing clients, the late-night share rises to 35 percent. Greenhouse, a larger applicant-tracking system, places roughly 15 to 20 percent of its voice-agent interviews after dark, according to Ophir Samson, the company's head of voice AI. The numbers, from different methodologies, point the same direction.
Asynchronous AI interviews let hourly workers, working parents, and people already in a day job record a take when their schedule opens up, instead of trying to take a Tuesday afternoon off for a video call. Ribbon cofounder Arsham Ghahramani frames it as widening the funnel for the workers least able to play the 9-to-5 game, and the 35 percent manufacturing share does suggest shift workers and second-job holders are using the late-night slot in real numbers.
There is no interviewer waiting to ask a follow-up, and there is no clock that ends the session at 5 pm. The candidate's living room is now the audition room, the camera is the gatekeeper, and the re-record button invites a small loop of self-corrections that can run until the candidate decides the take is good enough. The platform saves an hour of recruiter time per slot. The candidate spends that hour, or a version of it fragmented across re-takes and second-guessing, on their own kitchen-table clock.
The 30-seconds-to-think, 2-minutes-to-record, re-record-at-will format is not arbitrary. It is the product the platforms are selling to employers: a standardized, always-on screening step that does not require a recruiter to be awake. The trade is that the candidate's preparation, performance, and self-critique all happen without anyone on the other side absorbing the cost of that labor.
None of the platforms publish data on rejection rates by time of day, so the late-night slot cannot yet be tied to a worse or better candidate outcome. The percentages are also self-reported by executives with a commercial interest in the AI-interview category. The story is a platform claim about access, and Millard's rejection is a single anecdote, not a sample.
What the WIRED reporting does document is a measurable change in the geography of the interview: the candidate's home, the candidate's evening, the candidate's unpaid time. The test of whether the late-night wave is genuinely access, or a transfer of the scheduling problem onto the worker, will be whether the people using those slots (the Tim Millards, the shift workers, the second-job holders) start getting legible feedback on how the takes are graded, and whether the platforms publish any of it.
The next time a 9:30 pm email lands, the question is the same one candidates have been asking for years: who, exactly, is the flexibility for?