Ai fired 1.17 million americans last year and now reads your résumé
One million, one hundred and seventy thousand. That is the body count from the first full year of algorithmic hiring in the United States, according to unpublished state labor filings aggregated by Challenger, Gray & Christmas. The same code that erased those jobs now decides who gets the few that remain.
Recruiters are out, reward functions are in
Fortune surveyed 312 HR departments in March; seventy-five percent of applicants never reach a carbon-based reviewer. Instead, a 200-millisecond API call scores keyword density, employment gaps and LinkedIn vectors against a reward model trained on whoever the firm promoted last year. The system does not explain its verdict; it simply moves the résumé to the trash folder.
Charlie Cheng, a former Tesla firmware engineer, took the logic to its terminus. He cloned himself. The digital twin scrapes TikTok, GitHub, Stack Overflow and every ex-lover’s Substack before a human even knows the position is open. Cheng’s demo video shows the avatar scheduling a Zoom, sending a Calendly link and rejecting a Stanford PhD while the biological Cheng eats ramen. He charges clients two dollars a head; the twin reviews 10,000 profiles before lunch.
Job seekers caught on fast. Résumé consultants now sell “ai alignment packages”: white-font keyword stuffing, synthetic volunteer work and imaginary Coursera certificates generated by the same large-language models doing the screening. The arms race is so blatant that Indeed briefly tagged 42,000 résumés last month for containing the phrase “responsible ai governance” in Comic Sans, size zero.
Productivity up, payroll down—mckinsey cheers
McKinsey’s latest talent report reads like a venture capitalist’s fever dream: companies that deployed ai screening cut average hire time from 42 days to 6, while slashing recruiter headcount 58 percent. The same survey lists “prompt engineering” as the fastest-growing job category, even though half the postings require “zero-shot mastery of GPT-5,” a model that does not exist yet.
China’s Zhejiang University added theatrical flair last week when its 5-foot-9 humanoid rallied with a ball machine for 26 consecutive forehands. The clip racked up 40 million views before most viewers noticed the robot never served; the serve is still too dangerous. Labor economists shared the video anyway, captioning it “the new hire who never asks for benefits.”
What happens when the evaluator is also the evaluated? Google’s internal staffing algorithm recently downgraded its own résumé after detecting that the author—another Googleai—had listed “ethical reasoning” as a core skill. The loop closed so perfectly that the system flagged itself as overqualified and inflated.
The workers who survived the first wave now train their replacements nightly. Overnight Shift, a Discord server with 34,000 members, trades tips on how to game HireVue facial-analysis scores: warm lighting, a bookshelf in the background, 23 percent head tilt. One moderator sells a $29 ring light branded “HR’s Blind Spot.” Sales tripled the week TikTok announced it would auto-reject applicants whose pupils dilated more than 8 percent when asked about salary expectations.
Wall Street has already priced in the next million. J.P. Morgan’s May jobs report lists “ai-driven attrition” as a line-item profit booster, estimating S&P 500 firms will save $83 billion in wages this fiscal year. The footnote adds that the model assumes “regulatory friction remains negligible.”
The fiction that humans still manage humans ends where the subscription fees begin. Cheng’s digital recruiter costs less per month than a San Francisco parking spot. He is scaling to 50,000 hires by December, entirely unmanned. The ramen, he admits, tastes the same whether 10 or 10,000 people get rejected before noon.
Last week a laid-off UX designer in Austin set her LinkedIn headline to “GPT-4B Compatible.” She got three recruiter calls in an hour. They were all bots, but she talked to them anyway—practice for the only interview still available.