What Happened To The So-Called AI Job Apocalypse?

Authored by Joe Bertolami via RealClearMarkets,

A recent report from Stanford reviewed the latest employment data and found that, so far, AI has not resulted in large scale job destruction. Meanwhile, new hiring data from the Economic Times reveals that AI is actively fueling unprecedented job creation, with AI skills now powering nearly two-thirds of new Global Capability Center hiring. Together, these recent dispatches from the front lines of the labor market point to a calming reality: the much-dreaded AI job apocalypse hasn't materialized as a sudden extinction event.

The (sometimes buried) lede: AI is delivering real impact, and it is broadly changing the nature of work. But disruption is not a new phenomenon. The economy has always dismantled old work to build new work. What determines whether this evolution feels like progress or collapse isn't just the number of jobs lost, it's the speed at which that loss hits the labor market.

In 1995, Bill Gates circulated a memo titled "The Internet Tidal Wave," calling the web the most important computing development since the IBM PC. If the internet was a tidal wave, artificial intelligence is a tsunami. It is arguably the biggest advancement in computing since the Turing machine. Yet, from a distance, it's difficult to appreciate the speed of this wave, leading many to wonder when the broader economy will truly feel its impact.

To put this in context, we must understand the historical pattern already visible in the labor market. Combining decades of data from the U.S. Bureau of Labor Statistics and the Federal Reserve yields a remarkably consistent story of overlapping curves: job loss and job creation. Over the last two decades, nearly 20 million U.S. jobs vanished in disrupted sectors. Over the same period, total payrolls grew by 25.7 million. That equates to roughly 1.3 new jobs for every one destroyed. Classic examples include jobs in video rentals (-98.9%) and word processing (-83%) which largely vanished, but new work sprung up at the same time in areas like data processing (+54%) and warehousing (+260%) to support the digital economy.

The data also reveals an early signal that separates an absorbable decline from a brutal collapse: the disruption half-life, or how long an occupation takes to lose half its peak employment. Across the largest technological disruptions of the last few decades, the median half-life is about 10 years. Fast disruptions, like photo processing, take one to five years. Typical disruptions take eight to 13 years. And time is the ultimate shock absorber. When the economy transitions over ten years it feels like progress rather than a fast collapse, because it gives older workers time to retire and younger workers time to prepare.

If we track the most AI-exposed occupations-customer-service reps, IT support, telemarketers-since modern LLMs arrived in 2022, the early data is measured. After three years the current disruption looks closer to "typical" than a fast collapse, even before discounting the effects of offshoring, automation, and post-COVID corrections. This is Amara's Law playing out in real time: we tend to overestimate the effect of technology in the short run and underestimate it in the long run. The dire early warnings have given way to more cautious rhetoric. In 2025, Anthropic's Dario Amodei warned AI could erase half of entry-level white-collar jobs within five years. By 2026, he and OpenAI's Sam Altman are emphasizing productivity, economic growth, and the continued demand for human labor.

However, looking solely at total employment numbers masks a dangerous structural threat. Current evidence does not foretell the end of human labor, but AI is quietly breaking the mechanism by which we create experienced workers.

Software engineering is the canary in the coal mine. By most aggregate measures, employment looks stable; unemployment held at 4.2% in June 2026, and groups like the Yale Budget Lab find no clear AI effect yet on exposed occupations' absolute job totals. But the composition is shifting underneath our feet. Per AP and Oxford Economics, junior developer postings are down roughly 40% in four years. Employment for 22-to-27-year-old computer and math grads has fallen 8% since 2022, even as older grads in the same fields have edged up. This same erosion is surfacing wherever entry-level work once meant routine tasks: paralegals, junior analysts, and first-line support.

The paradox is that these industries keep growing even as their entry-level doors narrow. The BLS still projects software developers and QA analysts to grow 15% through 2034. But that projection relies on a pipeline that turns juniors into senior talent-precisely the pipeline now being choked off.

The reason lies in the nature of the work. Software development is a process of judgement and accountability: deciding what to build, executing it, and owning the result. AI is fluent at the middle layer-the well-specified, routine coding that once served as a junior's apprenticeship. But it remains far weaker at the judgment required on either side. The tasks AI automates are precisely the ones juniors were hired to learn on.

This is not merely an academic concern; it is a capital allocation problem. Misjudge the speed of disruption and you risk premature layoffs followed by a scramble to rehire, or funding the transition years too late, leaving you with a critical talent shortage when the leadership pipeline runs dry.

The challenge of the next decade isn't surviving the end of work. It is training the next generation of experts when the traditional paths to apprenticeship no longer exist. And businesses are beginning to realize this new reality as demand for AI continues to grow. IBM is tripling its entry-level hiring, redesigning those roles around the oversight of AI and systems thinking rather than cutting them. Rebuilding the entry-level on-ramp is now a competitive imperative.

Junior roles are not charity; they are talent capex. If AI creates more work than it destroys, companies will still need people who know how to run it, judge it, and fix it. AI may be the broadest technology yet, but that breadth is its best reason for optimism. A general-purpose technology seeds new work across every sector. The firms that recognize this, protect their entry-level pipelines, and keep training now are the ones who will own the senior labor market later.

Joe Bertolami is the Co-Founder and CTO at Clifton AI, an agentic context engine for investment research. Previously at Snap, Google, and Microsoft, with a couple of startups in between. He holds an M.B.A. from the University of Washington and likes using AI to write code, stories, and music, which he posts at https://www.bertolami.com.