New work still pays a premium, at least until the expertise spreads. A 2026 MIT study of US labor markets shows why young, highly educated workers are getting the first shot at emerging roles, and why startups should care.
The new paper from MIT labor economist David Autor and co-authors is not another abstract argument about whether AI will destroy jobs. It is a closer look at who actually gets pulled into newly created work, how much they earn, and what happens when once-scarce skills become common.
As MIT's Stone Center recently summarized, the study uses linked US Census data from 1940 and 1950, plus confidential American Community Survey data from 2011 through 2023, to track workers in job titles that did not exist in earlier periods. Between 2011 and 2023, 18.3 percent of US workers were in roles introduced since 1970. In 1950, 7.2 percent were in roles introduced after 1930.
The important finding for founders is not simply that new jobs appear. It is that new work pays more because it requires scarce expertise. That premium is roughly four times larger for technology-linked new work than for other forms of new work, and it lasts longer. In plain terms, the market pays up when a skill is useful, specialized, and still hard to find.
Who gets new work first
The demographic pattern is consistent across decades. Workers under 30 are most likely to hold new job titles, followed by those aged 30 to 54. Workers aged 55 to 64 are least likely. Education matters as well. In the study's preferred specification, workers with advanced degrees were 2.9 percentage points more likely than high school graduates to be employed in new work from 2011 through 2023.
There is also a clear geographic pattern. New work shows up disproportionately in urban areas, where dense labor markets make it easier for employers to find unfamiliar skill combinations and for workers to move into new roles. Once someone lands in new work, the advantage can compound. Workers employed in new work in 1940 were more likely to be in new work again in 1950, suggesting that early exposure to scarce expertise can keep paying off.
Demand shapes the jobs that appear
The study also makes a useful point for anyone watching industrial policy, AI infrastructure, or defense technology. New work does not emerge only because inventors create new tools. It can also come from demand. Counties that received new government-backed manufacturing plants during World War II generated more employment and more new work. The share of employment in new work was 0.35 percentage points higher in 1950 in those counties than in comparable places.
That matters because today's biggest demand signals are easy to spot. Clean energy subsidies, semiconductor investment, defense procurement, health care modernization, and AI data center buildouts are not just funding physical assets. They are creating demand for new kinds of operational knowledge. Startups that watch those flows closely may find labor market openings before larger companies formalize them into standard job descriptions.
What this means for AI and startups
The natural question is whether AI follows the same pattern. The answer is not automatic. AI can automate tasks inside existing jobs, but that is different from eliminating whole occupations. Most jobs are bundles of tasks, and when technology changes one part of the bundle, new responsibilities can appear around implementation, oversight, integration, compliance, and customer use.
For startups, the practical lesson is about training. New work does not arrive fully formed with a ready-made labor pool. It usually emerges through a mix of technology, demand, and deliberate investment in expertise. The companies that capture the early wage premium in AI-related work will often be those that build talent internally, rather than waiting for universities or large employers to define the field for them.
Health care is one place to watch. Public dollars account for a large share of US health spending, and AI could move in very different directions there. It could be used mainly to cut headcount, or it could help people with different levels of training take on more valuable tasks. The second path is harder, but it is also where new categories of work are more likely to form.
The premium fades, but not immediately
The study is clear that wage premiums decline as expertise spreads. Driving a car was once a scarce skill. Using WordPerfect once looked like a marketable advantage. The same will happen to many skills that now sound specialized, from prompt engineering to retrieval-augmented generation and fine-tuning workflows.
That does not make new work any less valuable. It means the window is finite. Startups need to either scale scarce expertise before it becomes routine or keep moving into the next layer of difficulty. The founders who understand where demand is forming, where skills are scarce, and how quickly those skills can be taught will have a real hiring advantage in the AI era. The rest will be competing after the premium has already started to disappear.
Also read: Clouted raises $7M to turn AI into a viral clip factory • Nvidia says it has 'largely conceded' China's AI chip market to Huawei • Anthropic's SpaceX compute pact shows how scarce AI capacity has become
