The Verge's Hayden Field tested three AI interview platforms — CodeSignal AI Interviewer, Humanly, and Eightfold — applying for roles modeled on her own job and real open listings at Vox Media. These aren't pilot programs. All three are live in corporate hiring pipelines now, which makes her first-person account useful in a way that a vendor demo never is.
The central pitch is scale plus neutrality: AI interviewers can screen every applicant rather than a small subset, and by removing human subjectivity, supposedly eliminate bias. Field's reporting challenges the second claim. Large language models are trained on internet-scale data that already encodes societal biases — the AI doesn't neutralize them, it just obscures who made the call.
On experience, all three platforms fell into the uncanny valley. None replicated the back-and-forth of a real conversation, and Field wanted a human on the other end every time. That's not just an experiential complaint — candidates who feel processed rather than evaluated don't always take the job.
For readers tracking the hiring-tech space: CodeSignal, Humanly, and Eightfold are all active commercial products with enterprise customers, and this is the most substantive mainstream audit any of them has received. The harder problem isn't bias in the abstract — it's that these vendors have built their pitches around a claim (neutrality) that's increasingly hard to defend. A company selling bias elimination has a more exposed position than one that never made that promise.