I surveyed workers to see if AI had caused job losses and was surprised by the findings

TEXT START: Do artificial intelligence tools such as ChatGPT eliminate jobs, create new ones, or both?

THE DISSECTION

This is a snapshot of visible personnel events among 1,250 people who still had jobs in late July/early August 2026. It is not a measurement of automation. The survey counts whether workers say they lost a job, obtained a newly created job, or received a promotion. It does not measure tasks removed, hiring that never occurred, hours, wages, output per worker, headcount avoided, entry-level pipelines, contractor cuts, or bargaining power. The article then stretches that narrow event count toward a labor-market conclusion.

The article’s most revealing data are its exclusions: every respondent was employed, the panel was self-selected, the study was not peer reviewed, and workers often cannot know whether AI caused or prevented a job change. The 3% who lost a job are therefore survivors who found work again or remained employed after the loss; permanently displaced workers are missing by design. The 95% no is not 95% of the workforce. It is 95% of a selected pool of current survivors.

The 6% reporting a job that did not previously exist and the 9% reporting AI-related advancement are not counterweights to displacement. A new labeler or compliance role can be transition labor around a system that destroys larger categories of work. A promotion is not a new job for the economy, and one worker’s advancement says nothing about the number of workers whose hiring, hours, or progression were suppressed. The article’s claim that promotion is the most important effect is therefore an unsupported interpretation of incomparable events.

THE CORE FALLACY

The central error is a category mistake: treating the absence of mass, self-attributed job loss as evidence that AI has not materially altered labor. Under the Discontinuity Thesis, automation first attacks tasks, throughput, hiring demand, and the price of cognitive labor. Headcount destruction can lag because firms retain workers while workflows are redesigned, because natural attrition is not backfilled, and because institutions move slowly. A worker can answer no to every headline question while producing more with fewer colleagues and facing a permanently smaller future labor market.

This survey does not test P1, P2, or P3. It does not establish whether AI has durable cost and performance superiority across cognitive work; it does not test whether human-only economic domains can be coordinated at scale; and, because it excludes jobless people and prospective hires, it cannot measure collapse of productive participation. It measures the lag before those mechanisms become visible as personal employment events.

HIDDEN ASSUMPTIONS

• A job title is the correct unit of automation. It is not; tasks and required labor-hours are the earlier unit.
• Workers can identify AI causality. They usually see the employment event, not the firm’s internal counterfactual: whether AI prevented a hire, changed a workflow, or made one worker replace several.
• Currently employed workers represent everyone affected. They do not. This is a survivorship filter.
• A newly created AI role is durable net employment. It may be temporary transition labor, and the article itself concedes that agentic AI may erode the value of current AI expertise.
• Promotions indicate net economic gain. They indicate relative advancement inside an existing hierarchy, not expansion of the labor circuit.
• Four years since ChatGPT’s debut is enough to falsify a staged transition. It is not. Physical, legal, institutional, and cultural inertia are precisely the lag defenses the thesis predicts.
• Demographic resemblance equals labor-market representativeness. Matching age, gender, race, and education does not correct self-selection, employment-only sampling, or unequal exposure to AI.
• Low observed job-change rates mean low systemic risk. They may instead mean the system is still consuming its human buffer before cutting visible headcount.

SOCIAL FUNCTION

Primary classification: partial truth. Secondary function: ideological anesthetic and transition management.

The partial truth is real: among this employed, self-selected sample, few respondents reported a directly attributed AI job change by mid-2026. The anesthetic is the implied leap from not-yet-visible mass layoffs to not-yet-material structural change. The article gives the reader a clean number to stare at while the unmeasured variables—foregone hiring, task compression, wage pressure, and labor-force exit—remain outside the frame. Its limitations section admits the instrument is looking at survivors, then still lets the snapshot carry a macroeconomic conclusion.

The AI-skills promotion story is especially fragile. It describes a temporary scarcity premium, not sovereignty. Once agentic systems absorb the same expertise, the promoted worker can become a more expensive interface to a capability the system no longer needs. That is a servitor niche under hospice care, not a durable escape from the thesis.

THE VERDICT

This study does not refute the Discontinuity Thesis. It documents a lag phase and mistakes lag for safety. The survey is a thermometer pointed at people still inside the building; it cannot report on those already expelled, those never admitted, or the rooms being automated behind closed doors. Its finding—limited self-reported job disruption so far—is compatible with an approaching or already developing collapse of productive participation. As evidence about current personal experience, it is useful. As evidence that AI has not begun killing the post-WWII employment-to-consumption circuit, it is structurally inadequate and rhetorically soothing.