Harvard Business School and the Burning Glass Institute wanted to know what happened after employers began removing degree requirements from job postings in 2021 and 2022. Roughly 85% of employers now say they have adopted skills-based hiring practices. But when researchers examined actual hiring outcomes, fewer than one in 700 hires could be directly attributed to removing degree requirements.
That’s not a rounding error. It’s evidence that hiring practices change much more slowly than hiring rhetoric. AI proficiency appears to be following the same path, only faster.
The demand is easy to see. AI skill requirements in entry-level job postings nearly tripled between fall 2025 and the spring 2026 hiring season, according to NACE’s Job Outlook 2026 survey. More than one-third of entry-level positions now require AI skills. Employers are writing AI expectations directly into job descriptions at a pace that outstrips almost every other emerging skill category.
The hiring process tells a different story. WGU’s Workforce Decoded report, based on responses from more than 3,000 hiring leaders, found that only 46% plan to expand skills-based hiring in 2026, let alone introduce AI-specific evaluation. Employers cited inadequate or expensive skills-testing platforms, while most applicant tracking systems still prioritize education, years of experience, and job titles over demonstrated capability.
The pattern is familiar. Employers increasingly say AI proficiency matters. Most still have no consistent way to evaluate it. The belief has changed. The hiring process hasn’t.
Three explanations are usually offered for this gap. Only one fully survives the evidence.
The first is inertia. Hiring processes are complex, and organizational change is slow. There is truth in that. But inertia explains a delay. It does not explain why organizations overwhelmingly describe themselves as skills-based while hiring outcomes remain largely unchanged.
The second explanation is infrastructure. WGU respondents did not question the value of AI skills. They questioned whether existing platforms could measure them efficiently or affordably, while most applicant tracking systems still rely on credentials and work history instead of demonstrated skills. Better technology will reduce part of this problem. It is unlikely to eliminate it.
The third explanation is measurement. It receives less attention, but the evidence suggests it is the most important.
AI proficiency is unusually difficult to evaluate because it does not fit the systems organizations have spent decades building. There is no universally accepted certification, no standard curriculum, and no consistent record of how the skill was acquired. Much of it is self-taught, developed through experimentation, and refined on the job. Separately from AI, 26% of employers already report that evaluating informal or self-taught skills is one of their biggest hiring challenges. AI proficiency may simply be the hardest version of that existing problem because the skill evolves faster than traditional credentials can document it.
That challenge extends well beyond AI itself. For decades, hiring has relied on credentials because they were easy to verify. AI disrupts that model. Much of the expertise employers now value is learned outside formal education, changes continuously, and leaves little traditional evidence behind. Hiring systems designed to evaluate yesterday’s signals are increasingly being asked to assess tomorrow’s capabilities.
The bottleneck isn’t adoption. It’s definition before measurement.
Organizations first have to decide what AI proficiency actually looks like in a given role. Only then can they measure it consistently.
One of our favorite partners, Canditech, is giving 50 talent acquisition teams a free, custom AI skills job simulation built around one of your open roles. You’ll work directly with their team, and it’ll be ready before the September hiring push.
The cost of getting this wrong is not abstract. Decades of industrial-organizational psychology research consistently show that direct demonstrations of relevant skills are stronger predictors of job performance than self-reported abilities or credentials alone. AI proficiency, when left unmeasured, defaults to the weakest available signal: a line on a résumé, a checked box on an application, or a claim made during an interview. For a skill this new and unevenly distributed, those signals are especially unreliable.
A longer-term view from McKinsey reinforces the same direction of travel. While NACE tracked a near tripling of entry-level AI skill requirements across two hiring seasons, McKinsey found that demand for AI fluency in U.S. job postings increased nearly sevenfold through mid-2025. AI fluency is now listed as a requirement in occupations employing roughly seven million workers. McKinsey notes that these figures reflect requirements appearing in job postings, not the actual skills of the people ultimately hired. Different datasets. Different time horizons. The conclusion is the same: employer demand for AI capability is accelerating far faster than most organizations’ ability to evaluate it.
Closing that gap starts well before selecting an assessment platform. Organizations first need to define what AI proficiency actually means for a specific role. The standard should be different for a financial analyst than for a marketing manager or a customer support lead. Knowing what AI can do is not the same as applying it effectively inside a real workflow. Without that distinction, hiring teams simply automate guesswork.
The employers most likely to close this gap will not necessarily be the ones with the largest technology budgets. They will be the ones that clearly define the capabilities they expect, identify observable behaviors that demonstrate those capabilities, and build those standards into a structured hiring process.
The interesting divide among employers is no longer between those who believe AI skills matter and those who do not. Nearly everyone has crossed that line. The real divide is between organizations that can demonstrate how they evaluate AI proficiency and those that simply list it as a job requirement.
That second group is still the overwhelming majority. It won’t stay that way.
The next phase of skills-based hiring won’t be defined by who values AI skills. It will be defined by who can measure them with evidence instead of assumption.
Hiring has never lacked belief. It has lacked evidence. AI is exposing the difference.
Sources
Harvard Business School & Burning Glass Institute. The Emerging Degree Reset: How the Shift to Skills-Based Hiring Holds the Keys to Growing the U.S. Workforce at a Time of Talent Shortage.
National Association of Colleges and Employers (NACE). Job Outlook 2026.
Western Governors University. Workforce Decoded 2026.
McKinsey Global Institute. AI: Work Partnerships Between People, Agents, and Robots.
Frank L. Schmidt & John E. Hunter. The Validity and Utility of Selection Methods in Personnel Psychology: Practical and Theoretical Implications of 85 Years of Research Findings.
