Economics

The Composition Problem: What the May Jobs Report Actually Shows About AI and Work

The May jobs report beat expectations at 172,000 jobs added. The sectors with the deepest AI deployment shed 35,000 jobs. The sectors least touched by AI hired 172,000. The productivity gains are real. The question is who receives them.

Mira Voss

Mira Voss

— 5 min read
A line chart showing aggregate job growth over sector-level divergence, with ledger-ink blue highlighting declining information and finance sectors beneath the reassuring headline number.
Original art by Felix Baron, Creative Director, Offworld News. AI-generated image.

By Duncan Galbraith | Economics | June 8, 2026

The headline number was good. The U.S. economy added 172,000 jobs in May 2026, nearly double the consensus forecast of 88,000. Unemployment held at 4.3 percent. Economists wrote about resilience. Markets responded positively. The story, as told by the headline, was reassuring.

Look at the composition and a different story emerges.

Where the Jobs Were — and Were Not

The May gains were concentrated in three sectors: leisure and hospitality (+70,000), local government (+55,000), and health care and social assistance (+47,200). Construction added 17,000. Warehousing added 6,000.

Two sectors declined: financial activities (-22,000) and information (-13,000).

This is worth sitting with. The sectors that grew — restaurants, bars, municipal services, hospitals, hotels — share a common characteristic: they are among the least automated sectors of the U.S. economy. AI tools exist in all of them, but at low penetration and without the structural capability to automate the core functions (a line cook, an ER nurse, a transit operator). The sectors that shrank — finance and information — are precisely the sectors where AI deployment has been deepest, most deliberate, and most extensively documented.

This is not a coincidence. It is a data point in a pattern that the headline number actively obscures.

The Bank Profits and the Bank Jobs

The financial activities decline (-22,000) arrived in a month when the economic backdrop for banks was benign. In the first quarter of 2026, the six largest U.S. banks — JPMorgan, Bank of America, Citigroup, Wells Fargo, Goldman Sachs, and Morgan Stanley — reported a combined profit of $47.3 billion, an 18 percent increase year over year. JPMorgan's Q1 net income alone was $16.5 billion.

Simultaneously, those same six banks collectively eliminated approximately 15,000 positions in Q1 2026. Citigroup is in the process of eliminating 20,000 roles by end of year. Morgan Stanley cut 2,500 globally. Bank of America's CEO Brian Moynihan explicitly attributed headcount reduction to AI, stating the bank eliminated work by "applying technology."

This is the mechanism in its clearest form: record profits, record AI investment, declining headcount. The productivity gains are real. The question is who receives them.

At the six largest banks, the answer is shareholders and executives. Not the workers whose job functions the AI replaced. Not the communities those jobs supported. The GDP contribution of financial services rises with profits. The employment contribution falls. The headline jobs number captures the latter but not the former; the profit figures capture the former but no one is printing them next to the payroll report.

The Information Sector: A Structural Decline, Not a Cyclical One

The information sector has now declined in four of the last five months per BLS establishment survey data. May's -13,000 follows a sustained pattern of reduction in a sector that encompasses software publishing, internet services, data processing, broadcasting, and telecommunications.

As of June 8, 2026, layoffs.fyi and comparable trackers document 183,966 technology sector job cuts across 247 events in 2026 alone — an average of approximately 1,157 job losses per day. Approximately 55 percent of these layoff events have explicitly attributed cuts to AI, automation, or machine learning.

The standard counter-argument is that tech employment is simultaneously being created in AI-adjacent roles. CompTIA projects a net 1.9 percent growth in the tech workforce for 2026, and May did see 69,000 tech occupation jobs added across the broader economy. But note the distinction between tech occupations (AI engineers at law firms, data scientists at healthcare systems) and tech industry employment (jobs at technology companies proper). The BLS information sector captures the latter. What it shows is contraction at the companies that produce AI — even as those companies report record revenues.

The tech industry has discovered, apparently to no one's surprise, that AI tools allow it to serve more customers with fewer software engineers, fewer technical writers, fewer content moderators, fewer QA testers. The productivity gains are real. They flow to the company's margins. The workers displaced are, by definition, not the ones collecting them.

The Wage Problem Has Not Resolved

Average hourly earnings rose 0.3 percent in May, bringing the annual figure to 3.4 percent. April CPI came in at 3.8 percent year-over-year, the highest since January. May CPI releases Wednesday, June 10.

The arithmetic: if May CPI prints anywhere near April's level, real wages remain negative. Workers are, on average, earning slightly less in inflation-adjusted terms than they were a year ago — for the third consecutive quarter.

The May jobs report's wage figure uses the all-workers average, which is pulled upward by high earners. Median wage growth — what a typical worker experiences — runs consistently below the mean. The sectors leading May's hiring gains offer substantially below-average wages: leisure and hospitality workers averaged $23.58 per hour in May, well below the economy-wide average of $37.53. The jobs being added are not replacing the jobs being lost on wages.

This is the distribution problem that the aggregate conceals: 172,000 jobs added is a real number, but it says nothing about the wage distribution of those jobs relative to the ones that disappeared. A software engineer earning $140,000 and a line cook earning $24,700 both count as one job. When AI displaces the engineer and the restaurant hires the cook, the employment number is unchanged. The income distribution is not.

The Productivity Thesis, Read Honestly

There is a version of the AI-and-labor story that goes like this: AI will generate productivity gains that raise wages and create new jobs at higher skill levels, as previous technological transformations did. This is the argument from electrification, from computerization, from the internet. The jobs of 2040 don't exist yet; the transition hurts but the destination is better.

This argument may be correct. The empirical record on general-purpose technologies offers reasons for long-run optimism that are not trivially dismissed. Acemoglu and Johnson's *Power and Progress* (MIT Press, 2023) provides the serious version of this debate — they do not deny that technological change can raise living standards, but they argue that whether it does depends entirely on who controls the direction of the technology and for whose benefit it is deployed. That question, they show, is political before it is economic.

The May payroll data is not a verdict on 2040. It is a snapshot of 2026, and what it shows is this: the sectors deploying AI at scale are cutting jobs while reporting record profits. The sectors unable to deploy AI at scale are hiring, at wages that run 37 percent below the economy average. The aggregate is fine. The distribution is not. The productivity gains are real. Their recipients are specific and narrow.

That is the story the headline number does not tell.

What to Watch Wednesday

The May CPI prints Wednesday, June 10. The FOMC meets June 16–17.

If May inflation holds near April's 3.8 percent, the Fed remains boxed — it cannot cut without appearing to abandon inflation control, and it cannot raise without compressing an economy already showing sectoral stress. The sectors most exposed to rate sensitivity (real estate, construction, capital-intensive tech) are carrying that constraint into a moment when AI is simultaneously compressing their labor demand. The margin of maneuver is narrow.

The jobs number was good. The question is good for whom, and what that means for Wednesday.

Sources

Read more