Nobody Works Here Anymore

This document references the GEM throughout. Here is how it is built, and why it reads the labor market at roughly five times the official distress rate.

The "no goddamned way these numbers are real" instinct is correct, and the agency itself has already conceded it, two years running, at record scale.

The economy added 818,000 fewer jobs in the 12 months through March 2024 than first reported, nearly 30 percent below the initial figure, the largest revision since 2009. Then they did it again, bigger. The preliminary benchmark revision for the 12 months through March 2025 was negative 911,000, the largest downward revision since at least 2000. Total nonfarm growth for 2025 got revised down to about 181,000 from 584,000, an average of roughly 15,000 jobs a month against a previously reported 49,000. The headline labor market was running about a third as strong as advertised, before each revision quietly corrected the record after the political moment had passed.

Even where the work exists, the path to it has been broken at two points before a worker ever reaches a hiring decision. First, an algorithm filters the resume out against criteria no applicant can see. Second, a large share of the postings were never real jobs to begin with. Both are documented, both are admitted by the employers running them.

The named, institutional source is the Harvard Business School and Accenture study "Hidden Workers: Untapped Talent," September 2021. Its findings are not soft.

The mechanism is mechanical and stupid by design. The software scans for keywords pulled from the job description, and a resume that does not carry the exact terms gets dropped regardless of whether the worker can do the job. Two of the most common knockout filters are an employment gap of six months or more and the absence of a four-year degree. The study found 49 percent of participating companies set their systems to eliminate any candidate with a gap of six months or longer, before any human judgment enters. The gap filter strips out caregivers, the recently ill, the formerly incarcerated, and anyone who survived a stretch of the labor market this document describes. And roughly 90 percent of executives admitted they already knew highly skilled candidates were being shut out, and kept running the systems anyway.

A machine rejects you in ten seconds for a job that, a meaningful share of the time, was never going to be filled by anyone.

The methodologies measure different things, intent versus outcome versus pipeline, and they converge on a 20 to 35 percent range. One in four to one in three postings is not a real door. The motives are documented and most are cynical. In a LiveCareer survey of 918 HR professionals, 45 percent admitted they "regularly" post ghost jobs and another 48 percent do it "occasionally"; 93 percent said their employer posts them. The stated reasons: pipeline building, keeping a board presence while not hiring, assessing the market, and projecting an image of growth so the company looks like it is not freezing hiring. A separate breakdown found 62 percent post fake roles to make existing employees feel replaceable.

The screening algorithm decides who gets in. A second AI process is deciding whether the door exists at all. Here the honest reporting has to hold two things at once, because the loud version and the quiet version point in different directions and only one is the real threat.

Challenger, Gray and Christmas, which tracks employer-stated reasons for job cuts, counted 54,836 layoffs in 2025 that companies directly attributed to AI. By early 2026 AI had become the single most-cited reason for cuts in consecutive months, accounting for roughly a quarter of all announced layoffs in April 2026: 21,490 of 83,387 cuts that month. Total announced cuts hit 1.17 million in 2025, the highest since the 2020 pandemic year. Workday cut about 1,750 jobs and pointed at AI reallocation; Amazon cut 14,000 corporate roles citing leaner AI-enabled structure; CEOs from Shopify to Duolingo have told staff in writing that headcount will not grow where AI can do the work.

This is the real one, and it does not show up as a layoff at all. The Stanford Digital Economy Lab study "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen (revised November 13, 2025), used high-frequency ADP payroll data covering millions of workers.

The headline finding: a 16 percent relative decline in employment for early-career workers aged 22 to 25 in the most AI-exposed occupations since ChatGPT shipped in late 2022, after controlling for firm-level shocks. Workers 30 and over in the same jobs held steady. That is the signature of AI eating the bottom rung rather than the whole ladder. The study found adjustment happens through employment, not wages, and is concentrated in roles where AI automates rather than augments human labor. The results are robust to excluding technology firms and remote-able jobs. Against it, employers wrote "AI" into job descriptions 400 percent more over two years.

Most Amazon packages are delivered by drivers who wear Amazon vests, drive Amazon-branded vans, follow routes assigned by an Amazon app down to the minute, and meet Amazon-set delivery quotas, while being employed, on paper, by one of more than 3,500 Delivery Service Partners. Those DSPs employ over 275,000 drivers. Amazon's position is that these are independent businesses and the drivers are not its employees, which is exactly how it keeps the workforce off its own books and out of reach of organizing.

That case is now being won, slowly, against Amazon. In Palmdale, California, about 84 drivers at a DSP called Battle Tested Strategies voted to unionize with Teamsters Local 396. An NLRB regional director found Amazon was a joint employer of those drivers because it controlled wage floors, routes, and schedules through its DSP contract. A second NLRB region, Atlanta, made the same finding at another DSP, MJB. Amazon yanked the Battle Tested Strategies contract in 2023 after the union vote, and in late December 2025 the Ninth Circuit refused to halt the NLRB proceedings. Amazon's response to losing the legal argument: arguing, in a separate federal filing, that the NLRB's very structure is unconstitutional. Faced with a finding that it actually employs the people who do its work, the company's move is to try to delete the agency that made the finding.

The personal-vehicle gig arm, where drivers use their own cars to deliver Prime Now and Amazon Fresh orders as pure independent contractors with no workers' comp and no benefits.

Amazon already paid the FTC 61.7 million dollars to settle charges that, between 2016 and 2019, it quietly cut Flex drivers' promised hourly rate and used customer tips to cover the difference, then concealed it when drivers noticed their pay falling. More than 15,000 Flex drivers have filed arbitration claims arguing they were misclassified and owed wages, overtime, and expense reimbursement. The Flex driver is the gig delivery driver wearing an Amazon badge: own car, own insurance gap, own depreciation, no employment.

Amazon built a model around intentionally high turnover. The average warehouse worker lasts about eight months. Jeff Bezos reportedly believed a long-tenured hourly workforce would become a "march to mediocrity," so churn was treated as a feature. A 2021 internal memo leaked to Recode warned that, on the existing burn rate, Amazon would deplete the available labor supply in its U.S. network by 2024, having already exhausted the worker pool in metros like Phoenix and California's Inland Empire. A meaningful share of warehouse staffing also runs through temp and seasonal agencies like Integrity Staffing rather than direct hire, and Amazon's holiday ramp routinely adds a quarter-million seasonal workers it does not commit to keeping.

In October 2025, The New York Times reported on internal Amazon strategy documents, presented to the board, laying out a plan to automate about 75 percent of operations and thereby avoid hiring workers it would otherwise need, even as it expects to sell roughly twice as much.

The projected savings: about 30 cents per item, an estimated 12.6 billion dollars between 2025 and 2027, and 2 to 4 billion dollars a year by Morgan Stanley's estimate. At the Shreveport, Louisiana test warehouse, a thousand robots already let the site run with about 25 percent fewer humans than it would otherwise need, with the design slated for 40 facilities by the end of 2027. The internal documents reportedly instructed staff to avoid the words "automation" and "AI" in favor of "advanced technology" and "cobots," the same instinct as the ghost-job posting and the ten-second AI rejection: do the thing, hide the name. Days later Amazon announced its largest corporate layoff ever, up to 30,000 roles. Bernie Sanders wrote Bezos directly asking what happens to the workers; MIT's Daron Acemoglu noted nobody else has Amazon's incentive to find the way to automate.

The gig sector is where the desperation gets monetized directly, and the headline framing of it as flexible entrepreneurship is the same survey-versus-records lie in a different costume.

The scale is no longer marginal. ADP's analysis of payroll data from more than 1.1 million employers found that individuals who received a short-term W-2 or a 1099 accounted for 27 percent of all jobs held in 2024. In a typical month about 1 in 10 workers were in the gig economy, but across the full year fully 1 in 4 had done gig work. Industry tracking puts somewhere around 59 to 70 million Americans doing freelance or contract work in 2025, roughly 36 percent of the workforce, with projections crossing half by 2027.

The security simply is not there. Only about 40 percent of independent workers have access to employer-sponsored medical insurance, 20 percent to life insurance, and just 5 percent to short-term disability. The Economic Policy Institute's national gig survey found 29 percent earned less than the state minimum wage that would apply under W-2 classification, 14 percent fell below the federal floor, 19 percent went hungry because they could not afford food, and 30 percent used SNAP in the prior month, roughly twice the comparable rate. The gig worker is the 1930s hobo with a car note: moving constantly to find work, except now burning personal vehicle equity to do it and paying for the privilege of being exploited.

That report works the dollar-count math honestly: the IRS 72.5-cent mileage cost that drops the median DoorDash hour below the federal minimum and into negative territory in several metros, the livery-exclusion insurance trap, the vehicle-as-informant telematics pipeline, and the wage-theft settlements already on the record in New York and D.C. It covers one facet. The structural point generalizes across the whole sector: the platform externalizes every cost of employment onto the worker, the worker absorbs the depreciation, the tax, the insurance gap, and the algorithmic firing risk, and the headline counts every one of those broken arrangements as a "job."

This is the through-line, and it runs from the construction laborer to the fastest-growing job in the country, the nurse. Companies do not hire en masse anymore. They rent labor through staffing agencies, and the contract between the agency and the client is the trap.

U.S. staffing firms employed roughly 2.2 million temporary and contract workers per week in late 2024, climbing through 2025 toward an ASA Index reading that translated to about 3.21 million weekly by mid-2025. LaborReady-style day-labor outfits put the construction laborer on a jobsite; BGSF and its peers put the maintenance tech in the building; travel and per-diem nursing agencies staff the hospital floors of the one sector BLS swears is booming. The "employer" on paper is the agency, not the company where the work happens.

The contract is where the worker gets eaten. When a client wants to convert a temp to a permanent hire, the agency charges a conversion fee. Staffing Industry Analysts puts the median temp-to-perm conversion rate among large buyers at just 10 percent in North America, meaning nine of ten temps never convert at all. Conversion fees commonly run 15 to 30 percent of salary, and many contracts only waive the fee after a stated hours threshold, frequently 480 to 720 hours. Read that threshold against a 90-day or 60-day assignment and the mechanism is obvious: the bounty is largest right up until the worker crosses the line that would make them free to hire, and the cheapest move for everyone except the worker is to cut them loose one day before that line.