Michael Strain Says the AI Boom Is Missing From the Data

Georgetown economist Michael Strain told a campus audience that the AI boom still does not show up in U.S. productivity data. The September 3 print put nonfarm output per hour up 1.4 percent in the second quarter, well short of the 5% to 6% trend he wants.

That gap is the live argument. Job-loss fear is high, campuses are buying tools, and investors still treat the technology as a new industrial age. The national accounts have not joined them.

Strain Tells Campus the Boom Is Not in the Data

Michael Strain is a professor of practice at Georgetown’s McCourt School of Public Policy and director of economic policy studies at the American Enterprise Institute. At a campus event covered on September 12, 2026, he said AI has not lifted productivity, has not moved the unemployment rate in a clear way, and has not remade daily life the way writing or the printing press did.

Investors and executives, he said, are still waiting for a gain the totals refuse to print.

When will it translate into productivity gains. First of all, it hasn’t yet, and I think there’s pretty broad agreement among economists on that point. There are some studies of specific firms or small groups of firms that do show that, but, when you look at the aggregate productivity data, it’s just not there.

Michael Strain, professor of practice, Georgetown campus event

He did not describe a jobless future. If prime-age work ever fell from 85 percent or 90 percent to 40 percent, he said, the country would face hunger in the streets or a much larger safety net. His worry is going the other way. “My level of concern is shrinking, not growing,” he said. “I really approached this question with a historical mindset, and concerns about the end of human work have been with us for a long time.”

He also argued that firms would not use the tools unless they cut production costs, and that lower costs would put downward pressure on prices. “AI will lower production costs, or it wouldn’t be used,” he said.

On hype, he would not give a clean verdict. “Is it overhyped? I think nobody really knows the answer to that question,” he said. Then he drew a line under the extreme claims: he does not think AI will take all the jobs, or kill everyone on Earth. “I think it will be a big deal.”

In March He Wanted 5% or 6% Growth

The campus line lands harder once it is set next to Strain’s own writing. In a March 26 essay he hoped for 5% or 6% productivity growth, a pace he said would beat the roughly 3% peak of the 1990s internet boom. At 3%, he wrote, living standards double in 24 years. At the 1.5% pace that followed the 2008 crash, they double in 47.

He was not cheering a slowdown. He asked how much disruption a rich country should accept in exchange for faster gains, then answered that the United States can handle a lot of it. Safety nets and schools, he wrote, leave advanced economies better placed than 18th-century Britain was for mechanized factories. “Creative destruction creates as well as destroys,” he wrote.

In early August he struck the same note in a column he posted himself: AI will quickly make most people better off, and over time the productivity it unlocks will lift wages and living standards. He warned policymakers not to hang new taxes on the technology to pre-pay for job losses that have not arrived.

On August 23 he quoted OpenAI chief executive Sam Altman saying everyone had been too ambitious on timelines, even with “this incredible technology.” Strain agreed on both halves of that sentence.

Sam Altman: “We’ve all been too ambitious on timelines, even with this incredible technology.”

He’s right that many technologist have been too ambitious on timelines. And he’s right that generative AI is amazing technology that will make all of society better off.…

— Michael R. Strain (@MichaelRStrain) August 23, 2026

So the September talk was not a conversion. It was a lag report. He still wants a boom. He is telling students it has not shown up in the totals, and that Silicon Valley’s civilization-scale version is the part he will not sign.

The Latest Productivity Print Is a 1.4% Gain

The Bureau of Labor Statistics revised second-quarter 2026 figures on September 3. In the nonfarm business sector, nonfarm productivity rose 1.4 percent at an annualized rate, as output rose 1.7 percent and hours worked rose 0.3 percent. From a year earlier, output per hour was up 2.2 percent.

That is not a collapse, and it is not the AI break Strain has said he wants. Through the second quarter, productivity in the current business cycle, which starts in the fourth quarter of 2019, has grown at a 2.1 percent annualized rate. That matches the long-term rate since the first quarter of 1947. It is faster than the 1.5 percent rate from the fourth quarter of 2007 through the fourth quarter of 2019. For 2025 as a whole, the annual gain was also 2.1 percent.

PRODUCTIVITY AGAINST THE BOOM STRAIN WANTS

Firm-level studies can still find a gain inside a company, which is the caveat Strain himself offered. The national series averages those wins with every office that bought a chatbot and then kept the same headcount and the same output. Until that mix moves, the boom is a local story.

The same release carried a quieter sting. The labor share, the slice of output that goes to workers as pay, was 52.8 percent in the second quarter, the lowest reading since the series began in the first quarter of 1947. Unit labor costs rose 1.2 percent as hourly compensation rose 2.6 percent. Workers are not taking a larger share of a faster-growing pie. They are taking a record-small share of a pie that is still growing at the old pace. The next productivity release, covering the third quarter, is due November 5, 2026.

Goldman’s 16,000-Job Monthly Drag

The job-loss number that still travels with this debate is older than the models now on Goldman’s own site. In 2023 the bank said generative AI could expose about 300 million jobs worldwide. That figure was about task exposure, not a forecast that 300 million people would be laid off. It is still the number people repeat.

By April 24, 2026, Goldman Sachs Research economist Elsie Peng was describing a much smaller realized hit. Weighing jobs where AI replaces people against jobs where it helps them, the team said AI had reduced monthly payroll growth by roughly 16,000 jobs over the prior year and raised the unemployment rate by 0.1 percentage point. In roles with more room to assist workers, the same work found about 9,000 extra jobs a month.

Peng noted that those estimates omit hiring for data-center construction and any extra labor demand from higher output, so the net drag may be smaller still. The losses that do show up, the team said, are falling mainly on younger, less-experienced workers. Telephone operators, insurance claims clerks, and bill collectors sat at the high-substitution end. Teachers, judges, and construction managers sat at the high-assistance end.

If replacement were already running at the scale the 2023 exposure figure implied, payrolls and output per hour would have moved together. They have not. The nearer tension is capital spending that still assumes a labor shock the monthly jobs report does not print.

THE AUGUST JOBS PRINT

• Payrolls: Total nonfarm employment rose by 162,000 in August 2026, the BLS said on September 4.
• Unemployment: The jobless rate was unchanged at 4.1 percent, with 7.0 million people unemployed.
• Pace: The August gain sat well above the 31,000 average monthly increase over the prior 12 months.
• Information: The information industry, the first place a software shock should appear, lost 23,000 jobs in August.

A 16,000-job modeled drag can hide inside a 162,000-job headline. It cannot hide a collapse in prime-age work. Strain’s 40 percent thought experiment remains a thought experiment.

73 Percent of Young Adults Expect Job Loss

The fear has not waited for the data. Pew Research Center surveyed 3,488 U.S. adults from June 22 to 28, 2026. 71 percent said AI will lead to fewer jobs in the United States over the next 20 years, up from 64 percent in 2024. Five percent said it would mean more jobs. Ten percent said it would not make much difference.

Among adults 18 to 29, 73 percent of adults under 30 now expect fewer jobs, up from 61 percent two years earlier. For the first time in Pew’s series, a majority of adults under 30, 55 percent, say they are more concerned than excited about AI in daily life. That share was 31 percent in 2021. Across all adults, 52 percent are more concerned than excited, up from 37 percent in 2021.

Those answers describe the next two decades, not August. They still shape what students ask a labor economist to address. Strain’s reply was that the end-of-work story is old, and that his own alarm is falling as the jobless rate holds near 4.1 percent. The survey and the campus talk are about the same technology. They are not about the same calendar.

Campus AI Programs Arrive Ahead of the Data

Georgetown is not waiting for a 5% trend either. In a February 23 letter, the university said it would put Google’s Gemini assistant in front of faculty and staff in early March, with students soon after, as the first personal generative tool it would offer. University Information Services now lets the community claim Google Gemini for faculty and students under an enterprise setup that, the school says, keeps chats and campus data out of model training.

On September 3, University President Eduardo Peñalver said Georgetown will write a campus-wide AI framework over two semesters: a set of core values by the end of fall 2026, and an operating framework by the end of spring 2027. He has also pushed a student AI Fellows program that pairs undergraduates with faculty to rebuild courses around the tools.

WHAT GEORGETOWN HAS STOOD UP

• Gemini access: An enterprise license for staff, faculty, and students, after a 2025 pilot, with a Google Workspace tie-in.
• AI Fellows: $2,500 for about 130 hours of student work; applications are due September 28, 2026.
• Rules on the way: Core values by the end of fall 2026, an operating framework by the end of spring 2027.
• Courses: A new AI ethics and application certificate in the College of Arts and Sciences, plus new AI coursework in the McDonough School of Business.

McCourt, Strain’s own school, named a fall 2026 Tech and Public Policy fellows cohort on August 31 focused on AI in government. Strain was also named this year to an AEI and Urban Institute commission on AI and the American workforce. The university is training people to govern a shock. Its labor economist is telling the same campus the shock is not in the production numbers yet.

Those two facts can sit together. A school can teach tools that have not yet moved GDP. The cost is confusion about what problem the new programs are for: classroom integrity, research method, or a labor market that still looks, at the headline level, like the one before ChatGPT.

The Printing Press Comparison He Will Not Stretch

Strain asked the audience to put AI next to the domestication of plants and animals, and next to the printing press. Those breaks, he said, changed where people lived, how they worked, and how politics ran. “It’s hard to imagine the American Revolution, for example, without the printing press,” he said. AI, in his ranking, “is going to be a big deal, but maybe not as big of a deal as some folks in Silicon Valley say.”

His March essay used a colder historical pair: Britain’s first industrial revolution, where typical wages stalled for decades, and the late-20th-century information boom, which he thinks the United States managed more cleanly. The open question he posed then is the same one he posed on campus. Will this wave look like crowded mill towns, or like the 1990s?

The data we have so far look like neither. Output per hour is advancing at the long-run 2.1 percent clip. Unemployment is 4.1 percent. A bank that once put 300 million jobs on a slide now counts a 16,000-job monthly drag. Young adults, by 73 percent, still expect the slide to win.

Strain’s second-order claim is that policy, hiring, and curriculum are being written for a labor crash and a productivity spike that the aggregates have not delivered. He still wants the spike. He will not pretend it is already here, and he will not treat the printing press as a usable forecast. The November 5 productivity release will be another test of that patience, not a verdict on the technology.