A few years ago, Ken Schumacher was working at a technology company. Part of his job involved evaluating potential hires: hopping on a Zoom call, giving the applicant an engineering test (a type of crossword puzzle with code instead of words), and continuing to “shut up for an hour” while the applicant struggled through it.
But most of the candidates had no problem. The company’s practices were being posted on sites such as Glassdoor. “All these 23-year-olds, of course, would practice the problem three times, come to me, and crush it,” Schumacher told me. “Now the big problem is everyone using AI to write their resumes.” They are also using chatbots and AI-powered teleprompters to help them reach the next round. It’s been “really, really hard for anyone to know who’s real and who’s fake,” Schumacher said. (This turned out to be his market opportunity – he now runs a startup AI to detect AI fraud and job candidates.)
The problem can be very serious in software engineering. But the same dystopian situation affects the entire labor market. AI has Tinderized hiring. Employees apply for hundreds of positions and never hear back; Companies receive thousands of resumes and struggle to respond. Additionally, AI has Amazoned rent. It has removed the exclusivity of software, flooded the market with similar versions, increased the number of frauds, and replaced personal choice with brutal algorithmic evaluation.
People were hoping that Silicon Valley could not only fix the equipment to get a job but also make the process fair. Unbiased tools would replace alma-mater networks. Digital websites can accept applications from anyone, anywhere. Employees could get free templates, practice exams and advice. “Technology generally tends to improve the efficiency of job matching,” Mitchell Hoffman, a labor economist at UC Santa Barbara, told me. But AI in particular seems to screw it up. Employers and employees are locked in an “arms race, where it’s AI-on-AI crime,” Kathleen Creel, a philosopher and computer scientist at Northeastern University, told me.
In just a few years, tools like ChatGPT and Claude have improved the creation of cover letters and resumes. Great stock of job applicants they use productive chatbots to polish their language and summarize their success—raising the average quality of these personal documents, at the expense of “compressing” and “enhancing” the information they convey, as one. Columbia Business School Paper to put it. In Silicon Valley, the phenomenon is sometimes called “signal collapse.” CVs were full of unexpected and unintended signs for hiring managers to analyse: degrees and certificates and languages spoken, as well as formatting errors and random errors and overly honest recruitment. Now everybody looks better and everybody looks the same and everybody uses keywords and everybody uses punchy action verbs. Hiring managers strive to “differentiate core expertise,” and separate signals of value from the chattering noise of chatter.
Because AI has reduced the time job seekers spend writing applications, people are submitting hundreds of them to ZipRecruiter, LinkedIn, and other sites—often for positions that don’t really fit, often to hear nothing back. Applicants do not get feedback on their strengths and weaknesses. They don’t understand what they need to do to get hired, or the skills they might need to get the job they want. Wave folding is carried out in both directions.
To sift through all those requests, companies are once again turning to AI. Resume Builder’s recent survey found that four out of five companies are using AI to analyze resumes, two out of five are using chatbots to communicate with candidates, and one in five are offering AI interviews. AI diagnostic tools are especially creative “Algorithmic Monoculture” in recruiting, a new survey of 4 million job applications has been found. Most candidates are rejected by every company they apply to; overall, hiring decisions are more accurate than they would be if HR managers didn’t use algorithmic screens. A decade or three ago, companies “would go through resumes and pick schools that they knew, or prestigious schools, or schools that had good programs” in the field, Creel told me. “That’s bad in its own right, from an equality perspective. But at least different people had different criteria for arbitrary investigation.”
Additionally, the screening rules discriminate against Black and Asian candidates, the paper found. “There’s this filtering that’s going on, and we don’t understand it, because these systems are not understood and are tailored to different institutions, and we don’t know the performance,” Sarah Bana, a co-author of the paper and a digital fellow at the Stanford Digital Economy Lab, told me. However, the tendency of the algorithm to benefit the advantaged and disadvantage the disadvantaged is evident.
For candidates who have passed the AI screen, a test like the one administered by Ken Schumacher can wait. Companies ask applicants to explain how they would respond to an unresponsive customer, or figure out an engineering solution to a software problem. In most cases, this step involves the AI writing the test, proctoring the test, and conducting the test. “I look at these things all day, every day, and I even have to pinch myself sometimes—what a fraud!” Schumacher said. “It’s absolutely ridiculous.”
The result is that AI is slowing down HR rather than speeding it up, because companies must spend more time reviewing applications and conducting background checks. A number of large companies are making face-to-face interaction a large part of the recruiting process. Google is to ensure that candidates have “at least one round” of face-to-face interviews, to “make sure the basics are in place,” Sundar Pichai, its CEO, said. Cisco is increasing “verification steps and enhanced background checks that may involve a personal component.” Companies are also growing trial periods and hiring people on a contract basis, with the intention of turning them into permanent employees after the supervisors have reviewed their work.
Finally, some companies use old-fashioned methods: referrals, alumni networks, internal job boards, headhunters. Schumacher said that companies were “returning to the clan.” They don’t do this because “Harvard engineers are the best in the world,” he said. “But it’s a much safer game in an age where it’s so hard to know who’s legit and who’s full of crap.”
Even with those methods, hiring has become more difficult—so much so that the entire labor market may become more difficult, and business transitions may fade slightly. Companies worry that employee retention will decrease, as managers end up having to fire unqualified employees and ineffective employees become frustrated and leave. Bana is concerned about businesses replicating their current workforce, and missing out on employees who can take them out of the group to think and expand their ambitions. Hoffman, a labor economist, raised the issue of AI reducing workers’ motivation to learn new skills and develop their expertise.
What’s a job seeker to do in this AI hellscape? No one I spoke to was sure. But being yourself in a cover letter and leaving it in person doesn’t seem like the worst idea.
