Written by Abhishek Shah
Every hiring team I talk to this year says some version of the same thing: they are drowning in applications, and a growing share of them look identical. AI writing tools didn’t just help candidates polish a resume. They flattened the differences between candidates almost entirely, and a growing body of research shows recruiters are struggling to tell who’s actually qualified from who’s just good at prompting.
At Testlify, we sit on the other side of that funnel, building assessments that companies use to figure out who can do the job rather than who can describe having done it. That vantage point has forced us to rethink what “fair and fast” hiring actually means, because right now most of what gets sold as AI-powered hiring is neither.
For years, speed and fairness pulled against each other. Move fast and you lean on shortcuts: a familiar pedigree, a keyword match, a gut call after thirty minutes. Slow down to be careful and you lose good candidates to a competitor who moved first. AI is supposed to have solved that trade-off. Mostly it hasn’t, because most vendors pointed the AI at the wrong end of the process.
Feeding a resume-screening model the same signals a tired recruiter already over-weighs (school name, past titles) doesn’t remove the bias in the system. It just launders it faster, with a confident-looking score attached. The useful version of AI in hiring sits upstream of the resume entirely: generating a real work sample at a volume no human panel could grade by hand, then scoring it consistently before anyone looks at a name or a school.
We ran that comparison for a client hiring backend engineers last year. The resume shortlist and the skills-test shortlist overlapped by less than half. A couple of candidates with unremarkable resumes, no name-brand employer, community college instead of a four-year degree, scored in the top of the group on the actual task. One of them is now running a migration project for that client. Nobody on the hiring team would have called that person in for round one on paper alone, and that gap is the entire argument for putting AI upstream instead of at the top of the funnel.
The speed gain is real, but it shows up somewhere most people don’t expect. A tool that just shaves a few days off scheduling calls isn’t the interesting part. The interesting part is a recruiter putting four hundred applicants through a real task in the time it used to take to phone-screen forty. That changes who gets a shot, not just how quickly the shot happens. A candidate buried on page six of an ATS ranking gets evaluated on the same footing as whoever sits at the top of the pile.
None of this works if the assessment itself is generic. We had to pull multiple-choice aptitude tests out of several clients’ hiring flows last year because candidates were answering them through an AI tool in seconds flat, and the scores stopped meaning anything. Banning the tools wasn’t the fix. We rebuilt the test instead, with open-ended tasks with deliberately messy requirements, the kind of ambiguity a real job throws at you, where an AI-assisted answer and a genuinely skilled one land in different places once a human looks at the reasoning behind them. NACE’s research on skills-based hiring points at the same pattern across employers broadly: the practice only works when the skill being tested is specific to the actual job, not a generic proxy for intelligence.
So before adopting any AI hiring tool this year, ask one blunt question before the sales demo even starts: does this change who gets evaluated, or does it just change how fast you evaluate the same narrow group you always would have anyway? Most of what’s on the market right now does the second. We only bothered building the first.
Author Bio:
Abhishek Shah is the founder and CEO of Testlify, a skills assessment platform that helps hiring teams evaluate candidates on real job tasks instead of resumes.