The unemployment rate appears to be in good shape. There are still job openings. On the surface, headline figures don’t seem to indicate a crisis. However, you get an entirely different impression if you speak with a payroll clerk who has seen her department quietly shrink over the past two years or a recent college graduate who is sending out applications month after month. It is getting more difficult to ignore the discrepancy between what the data indicates and what employees are truly going through.
The fact that AI isn’t changing employment isn’t the main issue. The reason for this is that America’s job loss metrics were never intended to capture this specific type of disruption.
Conventional layoff tracking looks for mass separations, such as workers filing for unemployment benefits, businesses announcing layoffs, and factories closing. The slower, more subdued erosion occurring in positions that just stop being posted, tasks that are taken over by software without any official announcement, or hiring freezes that never make headlines are all missed. According to a March Fox Business report, this was a “invisible layoff”—jobs were essentially eliminated through attrition and algorithmic replacement rather than pink slips. It’s possible that millions of jobs are vanishing in a way that traditional labor statistics were never designed to identify.
Using data from over a billion job postings, PwC’s 2026 AI Jobs Barometer provides one of the most insightful looks into this blind spot. It discovered transformation rather than widespread eradication. Strategic thinking, stakeholder management, and leadership judgment—skills that employees usually acquire after years on the job—are now seven times more likely to be needed for entry-level positions in highly AI-exposed fields. Since 2019, the number of traditional entry-level positions has decreased by 10%. The updated, revised versions increased by 35%. The position is on paper. All it takes is a resume, which no 22-year-old has had the opportunity to create yet.
This is referred to by PwC as “seniorization.” It’s a sanitized term for a painful truth. The ladder is still there, but it has been moved up a few rungs, making it inaccessible to those who most need it.
Software engineers are not the employees who operate most covertly. Payroll clerks, billing coordinators, customer service representatives, and HR administrators are a much larger, less visible group that economists have begun to highlight. Millions of them, dispersed throughout the economy, frequently work for mid-sized businesses in areas where tech layoff announcements are uncommon. These jobs, which are disproportionately held by women, may be similar to what deindustrialization did to working-class men in manufacturing towns a generation ago, according to New York Times reporting. That parallel is uncomfortable, and on purpose.
In their October 2025 analysis, the Yale Budget Lab at Brookings did discover something significant: the percentage of workers in high-AI-exposure jobs hasn’t decreased significantly yet. stability rather than collapse. However, they were cautious to point out that monitoring cannot end and that the data picture could change at any time. According to NBER estimates, there will be about 502,000 AI-related job losses in 2026, which is nine times the number from 2025. Whether something is moving is not the question. It’s whether the measuring devices are sensitive enough to detect it before it becomes obvious.
There is a feeling that the discussion has been reduced to a binary: either AI is causing a catastrophe or it isn’t, even though a more nuanced explanation would be more accurate. In order to avoid setting off alarms, work is being reorganized. Before there are training programs to replace them, skills are becoming outdated. Additionally, the employees who bear the initial costs are typically the least qualified to voice their opinions loudly.
It’s difficult not to feel that the problem’s invisibility is a contributing factor as you watch this develop. It’s simple to discount something when it doesn’t appear in a chart. However, this does not prove that the disruption is not genuine. It could simply indicate that the tools have not yet caught up.
