Artificial intelligence isn’t coming for construction workers, electricians, or factory operators. It’s coming for the office.
A new economic model from Anthropic — the company behind the Claude AI system — projects a future where blue-collar wages rise 5.9 percent above baseline by 2030, while white-collar knowledge workers see wages slip and employment in knowledge work fall 3.9 percent.
The reason? AI takes over the back-office tasks — engineering calculations, scheduling, administrative work — making physical projects profitable that weren’t before. Someone still has to pour the concrete, install the equipment, and keep the machinery running.
“SAVINGS IN THE BACK OFFICE BECOME OPPORTUNITIES ON THE SHOP FLOOR.”
Anthropic released the paper this week, modeling three possible economic futures through 2030. The “substantial adoption” scenario — the middle ground between mild change and science-fiction extremes — describes a powerful productivity and investment boom with plenty of work left for human beings.
By 2030, GDP sits 8.3 percent above its no-AI path. Annual growth reaches 5.4 percent, compared with two percent in the baseline. The capital stock — productive equipment and business assets — expands 13.8 percent.
AI could handle half of knowledge work by 2030 in this scenario, the model suggests. But most tasks still happen without it.
Here’s how the shift plays out: A company considers expanding a factory. AI makes the engineering, scheduling, and administrative work cheaper. That transforms a marginal project into a profitable one. The company greenlights the investment. That creates additional demand for labor — the building trades and the operators of the factory’s machines. Demand for skilled and unskilled manual labor expands even if those workers never touch AI directly.
Higher returns encourage more investment, which makes workers more productive. Real wages in occupations outside knowledge work rise 5.9 percent above their no-AI path. This broad category includes service workers and blue-collar workers.
Knowledge workers’ wages, however, fall 0.3 percent below their projected path. Employment in knowledge work drops 3.9 percent from mid-2026, largely because AI now handles so many of their tasks.
Overall unemployment reaches 4.6 percent, against a 3.8 percent baseline. While that’s a substantial increase — and if it happened quickly enough, it would set off recession signals — it would still be quite low by historical standards.
The model assumes wages adjust slowly, which is realistic. Employers avoid pay cuts that damage morale or drive away valued employees. Employees hate getting paid less for the same work. This is one reason why people lose their jobs in downturns rather than employers keeping the same workers on for less pay. The adjustment to lower demand tends to happen through layoffs.
A displaced accountant cannot become an electrician by changing his LinkedIn profile. Workers need to adjust their own expectations about what field they’ll work in and often face retraining costs. Unlike the pandemic lockdowns or a recession, many of the jobs AI displaces are likely gone forever.
But there’s an upside for white-collar workers. Total capital income is 18.9 percent above the no-AI baseline in 2030. Machines perform more tasks, increasing the share going to capital. Workers — especially knowledge workers — are also capital owners, typically in the form of retirement accounts and stock portfolios. Thanks to the Trump Accounts, many young people will become capital owners at a very young age.
The Federal Reserve’s 2022 Survey of Consumer Finances found that 78 percent of households between the 50th and 90th income percentiles owned stocks, directly or indirectly. Among the top tenth, ownership reached 95 percent. That means many professionals whose jobs are exposed to AI already have a financial interest in the businesses that benefit from it.
Stronger profits can support investment income and share values, cushioning weaker earnings for professional households. A larger retirement account can also reduce how much a family needs to save from each paycheck. For workers whose wages fall only slightly below their previous trajectory, that offset could be decisive.
Investment gains would likely cushion a significant part of the blow of job losses and transition costs for many established professionals, although young workers with little invested would remain more exposed.
The Federal Reserve is unlikely to be a passive observer while layoffs are happening. If productive capacity expands faster than spending, unemployment rises, and inflation weakens, the Fed could ease the stance of monetary policy to support demand. Faster productivity permits faster growth without necessarily creating inflation.
The displacement of white-collar workers may be milder than Anthropic imagines. The model already allows rising demand and new tasks to create work for people, but our economy may prove more resourceful than its assumptions suggest. The blue-collar workers with rising incomes will want financial advice, legal counsel, real estate agents, psychologists, and other white-collar services. As those services become cheaper, more households and businesses will be able to afford them.
We’re also likely to discover new occupations in which human expertise remains valuable, including some we would have trouble imagining today. Economies abhor unused potential, especially human potential.
The economic changes envisioned by Anthropic have important implications for immigration policy. Better technology and more capital allow a slowly growing workforce to produce substantially more. That means even with an aging population and a slow-growing workforce, we do not need supplementation from foreign workers to grow.
What’s more, mass immigration could do serious damage. The wage gains for non-cognitive workers in the substantial adoption scenario partly reflect their growing scarcity amid rising demand. Large inflows of competing workers could dilute that scarcity value and blunt their wage gains. Protecting those gains gives us a reason to restrain immigration even where hiring is strong.
The familiar plea for more “skilled” immigration crumbles in the substantial adoption scenario. For the most part, so-called skilled immigrants are cognitive workers. Computer-related occupations accounted for 64 percent of approved H-1B petition beneficiaries in fiscal 2024. With AI doing many of the tasks now performed by cognitive workers, adding skilled immigrants to the workforce will only exacerbate the downturn these workers face.
Immigration, skilled and unskilled, is likely to become largely obsolete as an economic matter.
For blue-collar Americans, the economic future sketched out by Anthropic is very appealing: more equipment to work with, more demand for their skills, and better pay. For white-collar Americans with savings, new jobs are likely to arise, and capital income is likely to substitute for diminished labor income.
Of course, some cynicism is probably warranted. The models were concocted by Anthropic, which has an obvious financial interest in pushing a positive story about AI’s effects on the economy. But the substantial scenario is plausible. And, frankly, it beats the dystopian predictions AI companies were peddling just a few years ago.
