Men Build AI, Women Get Replaced By It — The ILO's Two Labor Markets

The first near-term development worth watching closely is the EU Pay Transparency Directive entering real operation. EU member states faced a June 2026 transposition deadline, with companies employing 150 or more workers required to file their first gender pay gap reports in June 2027. Employers with unexplained gaps above 5 percent face mandated six-month correction windows and subsequent enforcement proceedings. I think this will produce a meaningful disclosure effect — the act of publishing a number that is publicly visible and comparable across competitors creates accountability dynamics that simply didn't exist before. Payscale data from U.S. states with well-designed pay transparency laws shows the controlled gender pay gap has been effectively eliminated. The six U.S. states where transparency laws haven't produced comparable results are the important caveat: legislative existence does not guarantee legislative effectiveness. The directive is a necessary condition for change, but whether it functions as designed will depend on implementation quality and enforcement seriousness.

The second near-term dynamic to track is the conversion of AI usage gaps into visible performance and compensation gaps inside large organizations. Survey data puts women's AI adoption somewhere between 32 and 35 percent, men's between 40 and 57 percent, depending on methodology. IMD-Wharton research associates emerging technology skills with a 6 percent salary premium — and McKinsey data shows women apply for emerging-tech roles at roughly half the rate of men. The structural setup for performance divergence is in place at organizations where AI tools have been deeply embedded into day-to-day workflows. I expect this gap to start appearing in performance review outcomes and bonus cycles within the next six months, particularly in finance, professional services, and technology companies where AI integration is most advanced. The magnitude doesn't need to be enormous for it to be directionally significant for career trajectories compounded over years.

A third short-term item that isn't getting sufficient analytical attention: the beginning of the first substantive wave of AI-driven workforce reduction. Through most of 2025 and early 2026, companies framed AI deployment primarily as augmentation — tools that help existing workers do more. As cost-reduction pressure intensifies through 2026's second half, the framing will shift toward substitution — tools that replace the functions of specific worker categories. When that shift happens in practice, the displacement concentration in administrative and office support roles will start showing up in actual workforce figures. The WEF finding that women hold 57 percent of AI-disrupted roles will start converting from a forecast into a reported pattern. I think the first clearly attributable large-scale examples will surface in media coverage between late 2026 and early 2027.

The final short-term signal to monitor is the BLS gender wage ratio trajectory. The fall from 83.9 percent in Q1 2025 to 80.6 percent in Q1 2026 was a 3.3-percentage-point decline in twelve months. The critical question is whether this represents a structural trend or statistical noise from sector composition changes. Q2 and Q3 2026 data will provide the first meaningful read. If the ratio holds below 81 percent for two consecutive quarters or declines further, it constitutes the first statistical evidence that AI-driven productivity gains are accelerating gender wage divergence in a measurable and directional way. If the ratio recovers above 83 percent, the Q1 2026 number likely reflects temporary sector-specific factors. I lean toward the former interpretation because the timing aligns precisely with the period of deep AI tool integration into white-collar workflows — and those tools appear to generate larger productivity gains in technical and analytical roles where men are overrepresented.

In the medium term — roughly six months to two years out — the most consequential variable is the pace at which national governments incorporate gender into their AI governance frameworks in a substantive and enforceable way. The current baseline is stark: 24 of 138 countries mention gender in their AI strategies, and only 18 include provisions with genuine policy teeth. If this count doesn't grow meaningfully toward 50 or more countries by 2027–2028, the structural AI gender gap will persist through active governance failure rather than technical inevitability. The EU is ahead of the curve on pay transparency but is not yet doing systematic AI policy work on gender-responsive automation management. Governments that design AI transition policy without gender lenses are effectively delegating the determination of who bears the disruption cost to market forces — and market forces don't self-correct for historically biased starting conditions.

Europe's tech sector gender ratio will be another pivotal medium-term indicator. The trajectory from 22 to 19 percent female representation in core tech roles is not merely a disappointing statistic — it is the leading demographic indicator of what the AI economy will structurally look like in 2030. The pipeline collapse — 20 percentage points fewer female STEM graduates converting to tech employment between 2023 and 2025 — needs to reverse within the next two years or the structural deficit becomes entrenched. The pipeline has two obstructions that need to be cleared simultaneously: the conversion barrier from STEM education to tech employment, and the advancement barrier from entry-level tech positions to management. Addressing only intake without addressing advancement produces a different ceiling geometry, not an absence of one.

The June 2027 first reporting cycle under the EU Pay Transparency Directive will function as a genuine social and commercial inflection point. When thousands of European employers simultaneously publish their gender pay gap data for the first time, it creates cross-company comparisons that have never existed in the public domain. Employers with large unexplained gaps will face talent acquisition disadvantages in a competitive skilled labor market — a market mechanism for voluntary remediation that operates independently of regulatory enforcement. This is a meaningful structural change in incentive architecture, not just a compliance exercise. My view is that it will produce real behavioral shifts among large employers, driven by commercial self-interest as much as principled commitment. The limitation remains: this addresses the accumulated historical wage gap, not the new AI-generated structural divergence forming in real time.

For BPO-dependent economies like the Philippines, the medium-term disruption will likely be faster and more visible than in advanced economies. With 12.7 million exposed jobs and Metro Manila's exposure at roughly 40 percent, the automation velocity of global corporations switching BPO contracts to AI platforms could outpace the national government's capacity to respond with retraining and redeployment programs. The social stakes are unusually high precisely because BPO has been a social mobility mechanism, not merely an employment category. When global firms replace voice agents and document processors with AI at scale, the first displaced workers will disproportionately be the young educated women who staffed those operations. I think major structural adjustment news from Southeast Asian BPO sectors will start emerging consistently in the second half of 2027.

A medium-term dynamic that's analytically underappreciated is what I'd call the AI ceiling — a new structural barrier that layers on top of the existing glass ceiling. Women currently face two simultaneous disadvantages in the same workplace: elevated risk that their existing roles will be automated, and lower organizational access to the AI tools that generate productivity-based career advantages. The second disadvantage means that women who successfully retain their positions through the initial automation wave may still fall behind male peers in performance metrics, because those peers have had more managerial encouragement and better organizational infrastructure to build AI fluency into their daily work. IMD-Wharton's 6 percent salary premium for emerging tech skills, compounded over a five-year career trajectory, produces a substantial earnings gap between two employees who started at the same level in the same organization. This is the AI ceiling: not a single barrier at the apex of a career ladder but a systematic tilt running through every rung of it.

Looking further out — two to five years ahead — McKinsey's 2019 estimate deserves revisiting with a post-generative-AI adjustment. The firm estimated that between 40 and 160 million women globally may face occupational transitions by 2030, up to a quarter of employed women worldwide. That estimate predates the generative AI inflection point by four years, meaning the actual transition pressure is plausibly higher. Critically, the sectors experiencing job losses are gender-differentiated: women's displacement concentrates in service and clerical work, while men's concentrates in manufacturing and mechanical trades. These are not symmetric shocks hitting a unified labor market — they are asymmetric shocks hitting gender-segregated segments of it. Policy frameworks that treat this as a single undifferentiated workforce disruption will produce gender-differentiated outcomes by default, regardless of whether that was intended.

The optimistic long-term scenario is genuinely viable, and I want to lay out its conditions clearly. WEF data shows women's AI engineering skills rising from 23.5 percent in 2018 to 29.4 percent in 2025, with the gap narrowing in 74 of 75 economies. If this trajectory holds, I see a plausible path to 40 percent female AI engineering representation by 2030 — not an official projection, but my conditional extrapolation from the current trend line. The share of women who believe AI will help their careers jumped from 47.2 percent to 58.8 percent in a year; attitudinal barriers are falling faster than structural ones. EU transparency requirements, ILO's gender-responsive AI design recommendations, and expanding U.S. pay transparency laws are simultaneously building institutional pressure from multiple directions. In the optimistic scenario, these converging pressures arrive at meaningful scale before the structural damage from the displacement wave becomes irreversible — and I genuinely hope that's where this goes.

For the optimistic scenario to materialize fully, three conditions need to hold simultaneously. Women's representation in AI development needs to grow substantially beyond 30 percent toward something approaching proportional participation. Organizational access to AI tools needs to be distributed without gender filtering — closing the manager-recommendation gap of 12 percentage points that currently operates from day one of a career. And national AI governance frameworks need to incorporate genuine gender-responsive provisions, growing from the current 18 countries toward something like 50 or more. These three conditions are mutually reinforcing: better representation in development produces less-biased systems, which produces less-biased deployment, which produces less-biased aggregate outcomes in employment and wages. Achieving one in isolation produces marginal improvement. Achieving all three simultaneously produces the structural change the scale of this problem actually requires.

The variable I believe is most consequential and least discussed is the gender distribution of AI venture capital allocation. Women-led teams currently receive less than 10 percent of AI venture investment. In a sector where $375 billion was invested in 2025 alone and $5 trillion in data center spending is projected by 2030, capital allocation determines the design direction of the technology itself. When more than 90 percent of AI investment flows to male-led teams, the systems that get funded and built predominantly reflect male assumptions about productivity, workflow, and the structure of work. Structural biases embedded at the architectural level — in training objectives, benchmark choices, application design priorities — are far harder to retrofit than surface-level algorithmic adjustments. The AI ceiling, the glass ceiling, and the displacement risk all trace back to the same source: the people at the decision-making table are overwhelmingly male, and that table's decisions shape everything downstream. This capital flow imbalance is, in my view, the most underestimated driver of the long-term outcome — and the least likely to correct itself without deliberate intervention.