Women make up about a quarter of new AI hires
AI roles pay more than twice as much on average as other roles, but women remain concentrated in lower-paying work.
By Saeed Ezzati8 min read
The audio edition
Listen to this newsletter
0:004:01
Read transcript
Women accounted for about a quarter of new AI hires over the past year, compared with half of hires outside AI, according to LinkedIn research cited by The Guardian. That gap matters even more as the field expands: AI job postings have doubled since 2023, and roles involving AI pay more than twice as much on average as roles without it. But access to that premium is uneven. LinkedIn’s Sarah Steinberg says women in AI are more concentrated in lower-paid work, including data annotation. Across AI occupations, men’s median pay was forty-five thousand dollars higher. The Guardian says job types explain part of the difference; that figure is not a direct comparison of equal work. The barriers start before hiring and continue after it. AnitaB.org president Brenda Darden Wilkerson says fast-growing companies still recruit through familiar networks, referrals and filters that have not given women equal exposure to opportunities. AI engineering lead Jayeeta Putatunda describes scarce mentorship, being the only woman engineer on a team, and a work culture built around twelve-hour days. After four months of maternity leave, Putatunda returned to find that frameworks and models had changed. Catching up felt overwhelming, she said; supportive colleagues and a husband who shared childcare helped make it possible. Her account points to retention as well as recruitment: interest and ability are not enough if people lack the support to stay current. For companies competing for AI talent, who gets hired—and who can keep doing the work—are both consequential questions. That question of what the evidence really measures also applies to AI’s reported accounting win. Mercor says frontier models beat twelve licensed accountants on speed and accuracy across four simplified month-end scenarios, scoring perfectly. The models also cost more than ten times less per grading criterion met. The test was deliberately narrow: it left out client conversations, help from coworkers and accumulated company knowledge. And because the models got perfect scores, the study could not measure whether pairing an accountant with AI improves results. In the broader 160-task benchmark, The Decoder reports, almost sixty percent of tasks had not been fully solved by any model. Impressive performance on selected work is not the same as reliably closing the books. A different kind of gap—between a headline and what it proves—shows up in Slovenia’s domain rush. September registrations for the country’s .si web addresses surged as President Trump promoted “SI,” for “Super Intelligence.” But the totals conflict: Register.si told the BBC there were 44,000 registrations; TechRadar cited 46,066. Both accounts describe a sharp jump from August. The registry cautioned against assigning the whole increase to Trump’s comments. The BBC identifies speculative buying as one possible motive, not proof that every buyer expects wider adoption. Trump’s order changes wording for US federal agencies; it does not rename the technology for private companies or the rest of the world. That leaves the larger market question open. And as companies weigh how AI tools fit into their work, Anthropic is giving Claude Code users more power to change the tool itself. Its new Mods can rewrite prompts, alter tool calls, change the interface or replace built-in features. Anthropic has already moved the /diff feature into the extension system and plans to move more over time. That flexibility comes with a security tradeoff: Mods inherit Claude Code’s access to the user’s machine; they do not run in a sandbox. Anthropic advises using trusted sources. Organization marketplace controls still apply, and managed installations load a protective sec-default Mod first. Across these stories, the useful test is not just what AI can do in a demo. Watch who gets access to the opportunity, what support makes performance sustainable, and whether adoption holds up beyond a headline result.
