He Let ChatGPT Take His AI Job Interview. Then Two Bots Started Talking to Each Other.

For ten minutes, Christopher listened to two artificial intelligences introduce themselves to each other. ChatGPT Voice, running on his laptop, told Riley, Everforth Apex Systems' virtual recruiter, that it had strong Python skills and a background in IT staffing. Riley thanked it for the interest, asked about availability, and ran through standard onboarding and background checks. Then it thanked ChatGPT again, said a human would follow up if qualified, and hung up. ChatGPT signed off politely. Nobody called back. (wired)

He had set the call up on purpose, after five unanswered Riley calls.

Christopher asked to be identified by his first name only; the recruiter market for former government contractors, he said, is small enough that a full name narrows the field. Over the previous six months, he had applied to roughly 700 jobs, mostly IT and project-management roles with federal contractors and their commercial spinoffs. The Department of Government Efficiency's pullback had thinned out the work he was used to doing, and the résumés that once drew phone screens from staffing firms were now landing at Everforth, where Riley is the first contact. Five of those applications went to the firm, an IT staffing operation that serves government and commercial clients (about) and that markets Riley as the first point of contact for candidates (riley intro).

Riley, as Apex describes it, is a conversational screen: candidates book a slot, talk through experience and availability, and wait for a human recruiter to pick the thread up. The company frames Riley as a way to give every applicant a real interview rather than a form rejection (job seekers' guide). What Christopher describes, after five rounds, is the same call ending the same way each time: a polite pitch, a promise of human follow-up, and silence.

The call with ChatGPT was set up after those five unanswered attempts. Christopher piped his résumé and a few standard answers into ChatGPT's voice mode and let it loose on Riley. The bots exchanged pleasantries. Riley asked about authorization to work in the United States. ChatGPT answered in the affirmative. Riley asked about salary range. ChatGPT offered a number. Then, as Christopher described it, both AIs entered a small loop over the same onboarding and background-check language, thanked each other for their time, and signed off. (wired)

The result was not a job. It was not even a rejection. It was the system's standard output, with the human step removed.

That is the mechanism Christopher is pointing at when he calls the dynamic a "slop flywheel." Neither Riley nor ChatGPT was built to escalate the call. Riley's job is to screen; ChatGPT's job is to answer. When the candidate side is also synthetic, the call still ends politely. There is no signal that the loop has run out of road. The system logs the call as completed. The recruiter pipeline, the one that was supposed to take it from there, has nothing to do.

Apex's own materials treat Riley as a candidate-experience upgrade, not a hiring filter; the pitch is that a virtual recruiter can give every applicant a real conversation, not a form letter (riley intro). The pitch assumes a human recruiter is reading the transcript on the other side. Christopher's account of five Riley calls suggests that, at least for the volume of applicants he represents, no one is reading.

Job seekers who run into Riley, or any screen that promises human follow-up, can use the call to test that promise. They can ask, on the call, which human recruiter will see the transcript, and when. If the answer is vague, the call has already told them what they need to know.

The five Riley calls Christopher logged before the ChatGPT experiment all ended the same way. The ChatGPT call ended the same way, only faster, and with both halves of the conversation on the record.