Hiring has quietly become a problem of volume. A single corporate role can draw thousands of applications, and entry-level roles in 2026 pull nearly three times the applicants of 2022. No human team can read all of it, so the work has moved to machines.
Researchers at Stanford University recently found that about 90 per cent of US employers use AI tools to sort and rank candidates. The efficiency is real. So is the temptation it brings, to treat speed as though it were judgment.
AI is very good at one thing. It studies past hiring decisions, learns which profiles succeeded, and ranks new applicants by how closely they resemble them. It optimises for what has already worked. But hiring is not a backwards-looking exercise.
Experience can be measured because it has happened. Potential cannot. A system trained only on the ‘past’ will, by design, keep selecting it.
The risk is not hypothetical. In 2018, Amazon abandoned an experimental recruiting tool after it taught itself to downgrade women, having learned from a decade of resumes from a male-dominated industry. No one wrote that bias into the code. The machine inherited it.
The same Stanford research points to a newer hazard. When many firms lean on the same screening algorithm, a candidate rejected by one is often rejected by all. One in ten people who applied to four such roles were rejected by all.
The people most likely to be filtered out are often the most interesting. A report by Harvard Business School found that 88 per cent of employers admit their automated systems reject qualified candidates simply because a resume misses a job description's exact wording.
The professional returning from a career break, the founder whose first venture failed but who now reads pressure better, the career changer whose skills are real but unfamiliar in wording, all tend to be screened out. Algorithms are built to recognise similarity. Potential usually shows up as a difference.
Work itself is being rewritten. The World Economic Forum's Future of Jobs report estimates that AI will create 170 million roles and displace 92 million by 2030. Skills that were central five years ago are already being redefined, and job descriptions cannot keep pace. A company hiring only against yesterday's template narrows its ability to adapt.
Some of the most important questions in hiring cannot be scored. Can a person sit with ambiguity? Do they learn quickly? Will they challenge a tired assumption rather than echo it?
These answers surface through conversation and context, not a ranking. Consider a leadership search down to two finalists. The first had the flawless profile: top institutions, well-known employers, and a clean and steady path. The second was harder to read, full of career shifts and choices that fit no standard box.
An algorithm scoring past data would have ranked the first well ahead. The second proved to be the stronger leader.
None of this is an argument against AI. Its value is real, freeing recruiters for the conversations that decide a hire. But its role is to assist judgment, not replace it.
The real advantage comes not from a sharper algorithm but from knowing where algorithms fall short. Teams built purely on pattern matching look consistent and think alike. Those who pair data with human insight can navigate change.
The future of hiring is not a contest between people and machines. It lies in understanding the limits of both. Algorithms can measure evidence. People recognise possibilities. Organisations are not built by choosing those who most resemble the past. They are built by recognising those who can shape a different future.
The author is founder and CEO of TalentiFi-X.
The views expressed in this article are solely those of the author and do not necessarily represent the views of THE WEEK.
