The economists who understand technology best cannot agree on whether AI is the sixth act in a 250-year story of progress, or the first act of something history has no name for yet.

It is the central fact of the labor market in 2026. On one side, Morgan Stanley published a careful analysis this week tracing five major innovation waves across 250 years.

Chief U.S. Economist Michael Gapen is explicit that those waves “are disruptive, capital-intensive and often volatile. They can displace workers, concentrate gains early and provoke political backlash. But over time, they raise productivity, restructure labor markets, expand output and, when institutions adapt, improve living standards broadly.”

The phrase “when institutions adapt” is doing enormous work in that sentence. The history bears the caveat out. Real wages in Britain barely moved during the first 80 years of industrialization, a period Acemoglu and Johnson document in Power and Progress as a sustained collapse in working conditions for the people inside the transition.

Child labor, 14-hour shifts, and unsafe factories were the daily reality, not a deviation from it. The right to organize, factory safety laws, and meaningful limits on working hours took another 50 years to win after that.

The “long run” in Morgan Stanley’s framing was, for actual workers, roughly a century and a half of paying the cost for gains that would eventually reach their great-grandchildren.

On the other side of the argument, three of the most influential labor economists alive, Daron Acemoglu, Simon Johnson, and David Autor, published a direct challenge to the historical optimism this week. Their argument is that “pure automation technologies” do the opposite of collaborating with workers. They “commodify human expertise, rendering it less valuable and potentially superfluous.” If they are right, the institutional adaptations that eventually rescued workers in previous waves may not have the same leverage this time.

The distinction Morgan Stanley draws between augmentation and substitution is real and matters for how you read the current data. Morgan Stanley Research Economist Diego Anzoategui says: “The same technology that automates tasks can also augment workers, increase productivity and boost demand in AI-exposed sectors. So far, the data suggest early, narrow displacement, more visible among younger workers, while overall disruption remains limited.”

That framing is accurate as a description of what the aggregate data currently shows. The Acemoglu et al paper does not dispute the current data. It disputes the inference that current patterns will hold as AI capabilities accelerate.

The historical record they rely on, the same record Morgan Stanley cites, suggests that even when the long-run outcome is positive, the transition itself can immiserate two or three generations of workers before institutions catch up.

The gender dimension of this debate has received far less attention than it deserves. A research brief published in March by the International Labour Organization, drawing on data from 436 occupations across dozens of countries, established a finding that changes the shape of the entire conversation.

Future Forwarded is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Female-dominated occupations are almost twice as likely to be exposed to generative AI as male-dominated ones. Around 29% of female-dominated occupations are exposed to generative AI, compared to just 16% of male-dominated occupations.

The difference is even starker at the highest automation risk: 16% of female-dominated occupations fall into the highest exposure categories, compared to only 3% of male-dominated ones.

The International Labor Organization (ILO) traces this gap to a structural feature of the labor market that predates AI entirely. Women are heavily concentrated in clerical, administrative, and business support roles, such as secretaries, receptionists, payroll clerks, and accounting assistants. Many of the tasks are routine and codifiable and therefore at higher risk of substitution by generative AI.

By contrast, men are more represented in construction, manufacturing, and manual trades, where tasks are less easily automated.

The problem compounds at the other end. The jobs being created by AI adoption, the ones carrying wage premiums and growing demand, are concentrated in engineering, cloud architecture, and AI development.

Globally, women accounted for only about 30% of the AI workforce in 2022, only 4 percentage points higher than in 2016. The displacement risk is concentrated where women work. The opportunity is concentrated where women are underrepresented.

That structural mismatch does not resolve itself through retraining programs: of the workers most at risk of losing their jobs due to AI, more than 6 million would likely struggle to cope because they are older, have limited savings, and face other barriers.

Most of those workers are in clerical and administrative jobs, roles that have historically been dominated by women.

The ILO’s senior economist Janine Berg put the core policy question plainly: The impact of generative AI on women’s jobs is not predetermined. “With the right policies, social dialogue and gender-responsive design, we can avert reinforcing existing discrimination,” she said.

If the policy choices are not made, the default outcome is an AI transition that widens the existing gender wage and employment gap.

The Morgan Stanley historical analysis and the Acemoglu et al warning are not actually incompatible once the timeline is named honestly. They operate at different scales.

Morgan Stanley’s data describes what aggregate labor markets are showing right now, and what historical patterns suggest about century-scale outcomes.

Acemoglu et al are describing the mechanism by which this wave could compress or extend the brutal middle period, the part where workers absorb the cost while waiting for institutions to catch up.

Both can be true simultaneously: the long-run historical record is broadly positive, and the transition itself can take three generations and require sustained political struggle to make those gains reach ordinary workers.

The American Enterprise Institute frames the burden of proof as belonging to what it calls “the discontinuity camp,” whose case “rests more on what AI could theoretically do than on what labor markets are actually showing.”

That framing is fair given the current aggregate data. The ILO gender numbers suggest, however, that for specific populations of workers, the discontinuity may already be arriving. Whether the institutional adaptations Morgan Stanley’s Gapen names actually happen in time, and for whom, is the political question that historical optimism cannot answer on its own.

A compilation of the Substack articles examining how the invasion already happened. You just weren’t invited. $9.95 flat fee for the bundle (PDF, ePUB), no subscription required. 2-hr reading time.

Get full access to Future Forwarded at futureforwarded.substack.com/subscribe

Get full access to The AI/Labor Report at ailabor.substack.com/subscribe