Fast Company reported this week that AI is burning women out at work faster than it's burning out men, and the data behind that claim is uncomfortably specific.
According to Fast Company's 2026 Workforce State of Mind report, 73% of women say mental or cognitive strain has hurt their productivity, compared with 67% of men. The gap widens on almost every downstream measure: 83% of women say the strain is disrupting their sleep, versus 70% of men. Focus takes a hit for 80% of women against 67% of men. Engagement drops for 69% of women versus 59% of men. These aren't rounding errors. They're a pattern, and the pattern has a mechanism.
The account was already overdrawn
Women spend roughly twice as many hours per week as men on childcare and household labor combined. That's not a new statistic, but it explains something new. Fast Company's reporting frames cognitive capacity as a bank account: AI adoption functions like a new monthly charge, the same amount billed to everyone. Men are frequently drawing from a surplus. Women are getting charged from an account that was already overdrawn before the first AI tool showed up in their workflow.
That imbalance shows up in how women are actually using, or not using, the tools available to them. Lean In's research found women are 22% less likely than men to be regular AI users at work. Only 30% of women said their manager actively encouraged them to use AI, compared with 37% of men. And women were 32% more likely than men to worry that using AI at work would make them look like they were cheating.
I'd argue that last number is the most telling one in the data. It's not a tooling problem. It's a permission problem. When the people handing out credit and cover are distributing both unevenly, "just use the AI tools" stops being useful advice.
The competence penalty nobody warned them about
There's a second mechanism at work, and it's uglier: women who use AI visibly pay a reputation tax that men don't. In studies of engineers using AI-assisted coding tools, women were roughly twice as likely as men to have their perceived competence downgraded, even when the code, the tools, and the output were identical. Reviewers didn't penalize the AI. They penalized the woman for having used it.
When the code and the tools are exactly the same, the only variable left is who gets believed.
That penalty produces a rational, if exhausting, response: quietly redo the AI's work by hand before anyone sees it, or avoid the tool in visible settings altogether. Either choice adds hours. Neither shows up on a productivity dashboard as "AI risk." It just shows up as burnout.
Layer on a third problem that hits everyone but lands hardest on whoever is already absorbing the most invisible work. Researchers at BetterUp and Stanford coined the term "workslop" for AI-generated output that looks finished but isn't. Their study found 40% of workers had received workslop from a colleague, and cleaning up each incident took an average of two hours. For a 10,000-person company, that pencils out to roughly $9 million a year in unbudgeted cleanup labor, and someone has to be the one who catches it, fixes it, and resends it without complaint. In most organizations, that someone skews female, for the same reason childcare and office housework have always skewed female: it's assumed, not assigned.
None of this means the tools are the problem. AI didn't invent the childcare gap, the credit gap, or the double standard applied to women's competence. It inherited them, then ran them at machine speed.
What actually helps
The fixes that show up in the research aren't abstract, and most of them are management decisions rather than personal productivity hacks:
- Managers explicitly encouraging AI use for women on their team, rather than assuming interest will follow once the tool is good enough
- Crediting AI-assisted work the same way regardless of who did it, with no quiet discount applied when the byline is a woman's
- Building workslop review into workflows before it reaches a colleague, instead of treating the two-hour cleanup as free labor
- Measuring cognitive load and sleep disruption, not just output, when evaluating whether an AI rollout is actually working
The organizations getting this right treat AI adoption as a change-management problem with an equity dimension, not a tooling rollout with a training video attached. That's a heavier lift than sending a Slack message with a new login link. It's also the only version of AI transformation that doesn't quietly transfer its costs onto the people already carrying the most.
