Your Open Roles Are Competing for a Shrinking Pool of Candidates
Sign Up For Daily Job Market News
For two years, jobs reports have stopped making intuitive sense, and Indeed Hiring Lab published the explanation this morning: the labor force itself is shrinking. The U.S. labor force has lost roughly 700,000 workers so far in 2026, and 973,000 since August of last year. Barring a surprise in the fourth quarter, 2026 will be only the fifth calendar year since 1948 with a shrinking labor force, and only the second time that’s happened outside a recession.
Payroll growth has still averaged 80,000 jobs a month this year, and unemployment fell from 4.3% in January to 4.1% in August. In July, the report initially showed an outright loss of 23,000 jobs alongside downward revisions, and unemployment ticked down anyway. In August, the reverse happened: a strong gain with upward revisions, and unemployment didn’t move at all. The mechanics are simple once you see them. When someone retires or stops looking for work, they leave the labor force calculation entirely, and the math around every headline number changes with them.
Indeed points to three forces feeding this. The first is immigration. Net international migration could fall to just 321,000 for the year ending June 2026, down from more than 2.7 million as recently as 2024. Foreign-born workers make up about 19% of the total labor force, but concentration matters more than the average here. Nearly a quarter of practicing physicians in the U.S. are foreign-born, along with 17% of nurses and nearly 40% of home health aides. Healthcare and social assistance job openings jumped 11% over the past year while hires fell 5%, a gap Indeed reads as a supply problem instead of cooling demand.
The second force is participation. Male labor force participation fell to 67.2% in August, down from 67.9% a year earlier and among the lowest readings since BLS began tracking the series in 1948. The third is the shift in who’s getting hired: non-farm jobs held by women rose 495,000 over the past year against 108,000 for men, putting women’s payroll jobs ahead of men’s for only the third time in U.S. history. Women accounted for 82% of all employment growth over the year, driven heavily by healthcare and social assistance.
Immigration and gender interact in a way that compounds the shortage. Foreign-born men participate in the labor force at 76.4%, more than 10 points above the 65.6% rate for native-born men. Slowing immigration removes a group of high-participation workers from the pipeline without any offsetting effect on the women’s side of the market.
Indeed’s bottom line is that the old rule of thumb, the labor market needing roughly 100,000 new jobs a month to hold unemployment steady, doesn’t hold anymore. This year, employers added roughly 80,000 a month and unemployment fell. In 2024, they added 122,000 a month and unemployment rose. Same rule, opposite readings, because the size of the workforce measuring against it has changed.
I’ve watched hiring managers use that 100,000 number as a mood ring for years, treating an 80,000 print as a warning sign and a 150,000 print as permission to expand. Healthcare openings up 11% while hires fell 5% is the cleanest version of the real story in this whole piece: more requisitions, fewer fills. Anyone staffing clinical roles has watched time-to-fill stretch for two years while the commentary insisted demand was cooling. Demand wasn’t the problem. The people weren’t there.
The immigration concentration point is the one worth watching most closely. Foreign-born workers make up 19% of the labor force overall and nearly 40% of home health aides, an occupation an aging country needs more of every year. Cutting net migration from 2.7 million to 321,000 while demand for that exact work keeps climbing is a math problem nobody in this piece has proposed a solution to, and it’s one every healthcare and long-term-care employer we work with should be planning around now, not after the next report confirms it.
Jobless Claims Stayed Near Multi-Year Lows This Week
The Department of Labor’s weekly claims report, released this morning, backs up the Indeed argument in the cleanest way possible. Initial claims for the week ending September 5 came in at 206,000 seasonally adjusted, down 1,000 from the prior week, against 259,000 a year ago. Continuing claims for the week ending August 29 sat at 1,774,000, versus 1,927,000 a year earlier, with the insured unemployment rate steady at 1.2%. No state triggered onto the Extended Benefits program this week.
Claims are sitting roughly 53,000 a week below where they stood a year ago, and continuing claims are down more than 150,000. Layoffs are not driving this slowdown. Read next to Indeed’s data, the picture holds together: payroll growth of 80,000 a month usually signals a weakening market, but claims this low say otherwise. Employers are struggling to fill open roles because the available workers don’t exist in the numbers needed. For anyone hiring right now, the pool of recently displaced candidates is thin and getting thinner, which means passive recruiting, going after people who aren’t actively job hunting, is close to the only recruiting that works in this market.
Staffing your team doesn’t have to be hard.
Reach out and see how we can help.
The More Comfortable Workers Get With AI, the More They Fear It
A new Gallup study adds a psychological layer to the same shortage: even in a market this tight, the workers using AI the most are the most afraid of losing their jobs to it. Workers who use AI daily or multiple times a week are more than twice as likely to fear their job will disappear within five years than workers who touch it only occasionally. This gap peaked in 2024 at 6.7 percentage points, and as of the first quarter of 2026, about 19% of all workers say AI will likely eliminate their job.
Gallup’s explanation comes down to one idea: the same tool that speeds up a task also raises the question of whether the person doing the task is becoming redundant. Workers who rarely touch AI never confront that question. Frequent users watch it move into their daily workflow, and without a manager explaining how that makes them more valuable, they start to feel expendable.
Management is the variable that changes the outcome, and the effect is large. Among frequent AI users, the link between heavy use and displacement fear is 6.8 points smaller for workers who feel respected at work, and 11.1 points smaller for workers who feel their organization cares about their well-being. Among infrequent users, management quality barely moves the number at all. Against a base rate of 19% expecting elimination, an 11.1-point swing covers a meaningful share of the total.
The costs show up in metrics employers already track. Workers who believe their job is at risk score 0.20 standard deviations lower on engagement, 0.19 lower on job satisfaction, 0.18 higher on burnout, and are 4.1 points more likely to be actively job hunting than the same worker reporting no such fear.
The 4.1-point figure is the one I’d sit with the longest. The employees most likely to believe AI is coming for their job are the ones using it every day, meaning your highest-adoption, most-adaptable people are the ones updating their resumes. Neither fix here costs a budget line. A manager sitting down with someone and saying what an AI rollout means for their role closes most of this gap, and most companies have avoided that conversation because they don’t know the answer yet. Silence gets interpreted, and it never gets interpreted generously.
What an AI-Heavy Future Could Do to the Job Market by 2030
Anthropic’s economics team published a modeled look at what AI could do to American employment by 2030, built on Department of Labor task data and released as an interactive scenario tool. It models three versions of the future: a modest scenario where AI lands roughly like the internet did, a substantial scenario where AI can perform half of all knowledge work by 2030, and an extreme scenario where AI outperforms people at nearly every knowledge task and creates almost no new knowledge work for humans.
Knowledge workers make up 62.2% of all U.S. workers today. In the substantial scenario, 59.7% are still in knowledge work by 2030, with 2.5% of all workers displaced and 1.8% landing in a new occupation. In the extreme scenario, that occupation-switching problem gets much larger, with a meaningful share of displaced workers still trying to find their footing when the model’s window closes in 2030. Anthropic names the transition directly: coders and call center agents shifting into work like electrician and nurse.
Pay moves in the same direction across all three scenarios, just at different speeds. Knowledge worker pay in the mildest scenario comes in roughly flat against a no-AI baseline, while every occupation outside knowledge work gains ground in every scenario, and gains the most in the extreme case. Anthropic’s own explanation ties this to a spillover effect: as AI speeds up knowledge work like design and permitting, demand for the physical labor those projects create, construction workers among them, rises with it.
Anthropic also surveyed more than 10,000 Americans in August with Morning Consult on what they expect AI to be capable of. The typical respondent’s answers land close to the substantial scenario, implying an unemployment rate around 5% by 2030. Anthropic is direct about the model’s limits: version 1.0 leaves out policy responses, business cycles, financial shocks, and any scenario involving highly capable robots, and it doesn’t track individual workers, so it can only give a coarse picture of what displacement costs any one person. Economists including Daron Acemoglu, David Autor, David Romer, and Jón Steinsson reviewed an early draft of the companion technical report, though Anthropic states they weren’t asked to endorse its conclusions.
I’ve watched people attempt far smaller occupational moves than coder to electrician, a financial analyst moving to operations, a recruiter moving to sales enablement, and those still take months and plenty of rejection inside the same broad skill set. Going from writing code to wiring a building is an apprenticeship, a license, and a pay cut on the way in before the numbers work again. Read this next to Indeed’s labor-shortage data, and the two are pulling in opposite directions on purpose. One says employers already can’t find enough people for the jobs open right now. The other says AI could eventually strand a large group of knowledge workers who can’t retrain fast enough to fill them. Both can be true at once, because the shortage and the surplus sit in different occupations, and the distance between those occupations is training time nobody has budgeted for yet.
