Women Are 26% of AI Hires, and the Window to Fix It Is Short

Women account for just 26% of AI hires, according to new LinkedIn research, even as those roles surge. A companion analysis found that millennials and Gen Z are the ones landing the fastest-growing, best-paid AI jobs.

The demand side of the AI boom looks broad. AI appears in 57% of Bay Area tech job postings, reporting out of Israel this week found demand for AI skills climbing fastest in nontech roles, and AI training is the single most planned corporate investment for the coming year. The supply side is much narrower.

We think that matters more than any single round of cuts this week, because the makeup of the AI workforce is being set right now, in a hiring window of two or three years. Whoever gets hired into these teams at the start is, a decade from now, who runs them.

AI teams recruit from a few feeder roles

Roughly a quarter of AI hires are women, which is below their share of technical hiring overall. We don't think the gap says much about who can learn the work. It mostly comes down to which existing jobs act as feeders.

AI teams recruit heavily from research, infrastructure, machine learning and platform engineering, the corners of tech where women were already underrepresented. They recruit much less from product operations, program management, design, QA, support and analytics, where more women work.

The AI shift runs in two directions at once. One set of jobs is being redesigned by AI, and another set of jobs is doing the redesigning. If you hire from a narrow feeder pool, the people most exposed to the redesign end up with the least say in how it gets specified.

Then it shows up in pay. Every credible read of the market says AI-adjacent work carries a wage premium, and a premium that attaches to a narrow group at intake doesn't correct itself. It compounds through promotion cycles, equity grants and the referral networks that fill the next round of jobs.

Age runs the same way. LinkedIn's data shows younger workers taking a disproportionate share of new AI roles. That usually gets told as a story about digital natives. The reasons look more ordinary to us: younger workers are cheaper to move around, less tied to a specialized legacy skill, and more likely to sit in jobs where employers will accept a six-month ramp.

Training can make it worse. Research on technostress covered this week found that badly designed AI training raises anxiety and lowers confidence, particularly for workers who already fear becoming obsolete, the same worry HR Dive has been reporting under the name FOBO. When companies call AI training their top planned investment for the next year, our guess is that most of them are buying uniform courses priced per seat. That's the format most likely to widen the confidence gap it's supposed to close.

Hidden skills send employers back to old proxies

Half of Gen Z workers say they feel guilty using AI at work, and four in ten hide that use from their employers. That looks like a culture problem. We'd call it a measurement problem. If a big share of the workforce is building real fluency in private, employers have no reliable way to see who can actually do the work.

Without that, hiring falls back on proxies. Inc. reported this week that the strongest candidate for a technical role often isn't a computer science major, and that's especially true of AI-adjacent work, where domain knowledge plus tooling often beats a traditional degree. The research software engineers at the John Innes Centre, who bridge biology and computer science, are a good example. But skills-based hiring only works when skills are visible. When they aren't, employers go back to the oldest proxies they have, which are prior title, a prestigious employer, a degree, and age.

That's the link between the guilt number and the 26% number. Private practice plus proxy-driven hiring gives you a pipeline that looks like the last decade instead of the next one.

Governments are moving, mostly in a sensible direction. Singapore's SCS has launched an AI skills pathway, and IMDA introduced AI fluency programmes for the legal sector, a deliberate push of AI skills into professions outside engineering. New York is holding listening sessions on AI job disruption before it writes rules. The New York Times profiled an Indian city where AI work is creating jobs for people rather than deleting them.

Almost none of these programs look at who enrolls. They add to the total number of trained workers without changing the mix of people hired into the best-paid tier. Other forces push the opposite way: visa holders are leaving Big Tech over layoff fears, and workers in places like Oklahoma are heading into the trades instead of competing for AI-adjacent roles.

The risk we see over the next five years is a labor market with two tracks. One has AI-native jobs with premium pay, concentrated in a few metros and a narrow slice of workers. The other has AI-adjacent jobs that take on the tools without the raise.

There are openings, though. The World Economic Forum documented this week that the path from junior to senior roles is being rebuilt, and a rebuild is when the makeup of a workforce can change. Smaller firms also say the layoff wave has eased their hiring, so employers outside the headlines can now reach people they couldn't afford before.

Those are the places we'd look first. Non-tech employers, where AI skill demand is rising fastest and feeder roles are broader. Mid-size and smaller firms, which filter less rigidly on credentials and can now get senior people who weren't available in 2023. Hybrid roles in law, biology, construction and finance, where the scarce input is domain knowledge rather than model expertise. And the cheapest route of all into an AI team, which is a lateral move inside a company that already knows your work.

If you're not in the 26%, stop hiding your AI use. The guilt number is the one you can act on today. Write down what you built with AI, what the tool got wrong and what judgment you applied, and you become visible to the people who make promotion decisions. Then treat your distance from the feeder roles as something you can change. A six-month rotation onto an evaluation, data pipeline or deployment team will change your next recruiter screen more than another course completion will.

If you work in a nontechnical function, the surge in AI demand outside tech is your advantage. Domain expertise plus demonstrated tooling fluency is underpriced right now compared with generic AI skills, and it's the combination we'd bet on surviving the next redesign.

For managers, the 26% figure isn't something to hand off to universities. It gets decided role by role, in the next hiring cycle, and it will be far more expensive to fix in 2031.