A Year After the "Half of Entry-Level Jobs" Forecast, the Bar Moved Instead of the Headcount

The mass displacement event that AI executives forecast last year has not arrived, and the CEOs who forecast it are softening their language toward augmentation rather than replacement ^[1][2][6]. For anyone holding a workforce plan built on the elimination scenario, the operative risk has flipped: the cost of over-cutting now exceeds the cost of over-hiring, and the real change is showing up in job specifications rather than in headcount.

The original claim was specific. Anthropic CEO Dario Amodei said in May 2025 that "half" of all entry-level white collar jobs would vanish, and a month later OpenAI's Sam Altman predicted the end of "certain job categories" ^[1][2]. Companies started naming AI in their layoff notices, workers organised, and students changed their career plans ^[3]. A year on, the carnage has not appeared, and economists note that AI capability has advanced far faster than the economic transformation, much as in previous technology cycles ^[4][5].

The numbers do not support a displacement story. According to an analysis from the Stanford Institute for Economic Policy Research, unemployment among the 20% of workers most exposed to AI rose by 0.77 percentage points since 2022, when ChatGPT launched, compared with 0.85 percentage points for the least exposed ^[7]. That is a 0.08 percentage point gap in favour of the group supposedly most at risk ^[8]. Recent graduate unemployment did hit 5.6% against a national average of 4.2% earlier this year, roughly 1.4 percentage points higher, and the report allows AI as a possible factor while also pointing to remote work and the unwinding of pandemic-era overhiring ^[9][10][11]. Erika McEntarfer, a fellow at the institute and report co-author, said employment trends in the occupations where impacts would appear first are "largely stable" and that the pattern resembles the computer revolution, which took decades to reshape labour markets ^[12].

Caveat the data honestly: government statistics lag and do not track specific technologies, while private data is more current but less comprehensive, so economists agree there will be an effect but cannot size or time it ^[13].

What is moving is the entry requirement. ZipRecruiter's latest employer survey puts 74% of employers treating AI skills as a strong advantage or a requirement, with 13% requiring them company-wide rather than only in technical roles, and half expecting candidates to arrive as practical or advanced AI users on day one ^[14][15]. Those demands do not always appear as the word "AI" in a posting; they surface as higher expectations on speed, quality and self-sufficiency ^[16]. ZipRecruiter labour economist Nicole Bachaud describes "a rising bar rather than a shrinking pool," with the worker problem being skills-matching more than job scarcity ^[17]. Employers are adding and cutting inside the same functions, including tech, customer support, and business management and operations, which Bachaud reads as evidence they are still working out the skill set that succeeds ^[18]. Stanford's Nicholas Bloom calls this turbulence: some jobs destroyed, others created to implement, sell, fix and develop AI systems ^[19]. NYU Stern's Robert Seamans, a co-developer of a standard measure of occupational AI exposure, sorts the effect into jobs made obsolete, jobs created, and jobs changed, and says the third bucket is by far the biggest ^[20].

Two things to watch. Some economists expect AI to push more work toward freelance and contract arrangements as companies determine which skills they actually need, which would show up in contingent-labour spend before it shows up in unemployment ^[21]. And watch whether your own postings are quietly raising the bar on speed and self-sufficiency without anyone deciding to do so ^[16].