Goldman Sachs economists previously estimated that 6% to 7% of US workers, roughly 11 million people, face imminent AI displacement. But the horror show doesn’t end with the layoff notice. According to Goldman’s analysis of 40 years of labor data, workers displaced by technology suffer a “scarring effect” that lasts decades.

The numbers are brutal: tech-displaced workers take an immediate 3% hit to real earnings compared to those displaced for other reasons. Ten years later, their earnings remain 10 percentage points below workers who never lost their jobs. They face elevated unemployment risk for a full decade and experience delayed homeownership and slower wealth accumulation throughout their careers.

The mechanism is occupational downgrading. Workers displaced by AI are more likely to move into routine occupations requiring fewer analytical and interpersonal skills, the same technological shifts that eliminated their positions also eroded the market value of their expertise. Real-world case studies of large-scale human-to-robot displacement demonstrate this isn’t theoretical, Amazon’s automation roadmap targets 600,000 US workers while carefully avoiding words like “layoffs” in favor of corporate euphemisms.

On the flip side of the paradox sits the intensity engine. Bloomberg’s April 2026 cover story documented the “Great Productivity Panic of 2026”, where AI coding agents like Claude Code and OpenAI’s Codex haven’t freed engineers, they’ve chained them to their desks longer.

UC Berkeley researchers spent eight months observing employees at a 200-person tech company and found a counterintuitive pattern: workers using AI increased both the volume and variety of work completed, but also took on significantly more total work. As one engineer put it: “You had thought that maybe, ‘Oh, because you could be more productive with AI, then you save some time, you can work less.’ But then really, you don’t work less.”

The research shows hours expanded to fill the available capacity. Google’s DevOps Research and Assessment (DORA) team surveyed nearly 5,000 technology professionals and found that while 90% use AI at work and over 80% report productivity gains, they also experienced increased software delivery instability, more frequent code rollbacks and patches after release. More code shipped, more problems created, more hours spent fixing the mess.

A data scientist writing on Reddit captured the reality beneath the “vibe coding” hype: “It is now 90% debugging, 10% coding instead of 10% debugging, 90% coding.” The tools create autonomous-seeming systems that require constant human correction, shifting the work burden from creation to remediation without reducing total labor time.

The paradox isn’t a bug, it’s a feature of late-stage capitalism’s relationship with productivity. As one Reddit commenter noted, there’s a fundamental misunderstanding of corporate incentives in the “AI will free us” narrative: “People are weirdly underestimating corporate greed when they spin up AI labor market dystopia takes. It usually amounts to assuming there’s some satiation point… But bosses are greedy pigs! You think they’re going to be happy with cutting 90% of headcount to get today’s profits when they can just work their current workforce to death and get even more?”

This creates the demand destruction problem that economists are quietly panicking about. Downstream economic consequences of automating customer demand create a feedback loop: if AI replaces a large number of jobs, disposable income vanishes. If consumer demand drops, the revenue supporting those AI efficiencies collapses. Yet individual firms face a prisoner’s dilemma, any company not cutting costs through AI falls behind immediately, even if the collective outcome destroys the consumer base.

The ADP Research Today at Work 2026 report, based on survey responses from more than 39,000 workers across 36 countries, found that daily AI users were four times as likely as non-users to say they were not as productive as they could be. The explanation is psychological: AI has automated the checklist-style tasks, emails, summaries, first drafts, that gave workers tangible daily accomplishments. Without those small wins, workers feel they’ve done less even when objective output has increased.

Omar Abbosh, CEO of Pearson, argues the AI job apocalypse is largely a Silicon Valley story. Software engineering is one of the first professions where AI delivered real, visible productivity gains, so anxiety travels fast from that sector. But extrapolating from tech to the entire economy is dangerous.

Oxford Economics found the evidence of an AI-driven shakeup to be patchy at best, and Wharton School researchers cite “AI-washing” of job losses that actually reflect economic cycles and pandemic over-hiring. The US unemployment rate sits at 4.4%, far below EU peaks in the 1990s when rates hit 11% overall and exceeded 20% for young workers.

McKinsey data suggests two-thirds of companies using AI have not scaled it across their enterprise. The diffusion is slow because the hardest problems aren’t technological, they’re organizational. Data readiness, security, integrations, workflow redesign, and building human skills remain stubborn bottlenecks.

Yet this comfort is cold. Debunking the correlation between AI adoption and actual workforce reductions reveals that many current layoffs reflect pandemic hiring bloat rather than algorithmic substitution, but that doesn’t mean the substitution isn’t coming.

This phenomenon has historical precedent. Email made communication instantaneous but created constant inbox pressure. Smartphones enabled remote work but dissolved boundaries between work and personal life. AI follows the same pattern: technological tools make work faster, but instead of reducing working hours, they lead to higher expectations and increased workloads.

Harvard Business Review even coined the term “AI brain fry” to describe the information and workload fatigue that comes with too many AI tools operating simultaneously. Automation increases intensity and stress rather than productivity when poorly implemented.

The ManpowerGroup 2026 Global Talent Barometer found that while regular AI use increased 13% in 2025, confidence in the technology’s utility fell 18%. More adoption, less trust. Workers experience FOBO, Fear of Becoming Obsolete, a creeping sense that skills are degrading in real time while the window to stay relevant closes faster than they can adapt.

There is one bright spot in Goldman’s otherwise bleak analysis: retraining works. Workers who retrained after a tech-driven job loss saw an average 2 percentage point increase in cumulative real wage growth over the next 10 years. Their probability of being unemployed declined by around 10 percentage points.

Retrained workers tend to move up the occupational ladder into roles with higher abstract content, positions requiring advanced skills and greater complementarity with information and communication technology. The catch? Someone has to pay for it, and companies will never willingly help pay for universal transition programs when quarterly earnings demand immediate cost extraction.

The AI economic paradox presents two equally unpleasant futures. In the first, you’re displaced, earning 10% less than your peers a decade later, having downgraded from analytical work to routine tasks, your wealth accumulation permanently stunted. In the second, you keep your job but work longer hours under higher expectations, debugging AI hallucinations instead of writing code, suffering “AI brain fry” while your employer extracts every efficiency gain as additional output rather than reduced workload.

The uncomfortable truth is that AI isn’t replacing work or creating more work, it’s reshaping where effort goes, and current institutional incentives ensure that reshaping benefits capital over labor. Until organizations deliberately use productivity gains to reduce working hours rather than increase output, the paradox remains: you’ll either be automated into economic precarity or intensified into burnout, with no option C in sight.

The companies that win won’t be those that implement AI fastest. They’ll be the ones that implement learning fastest, both machine and human. But given that most organizations are reinvesting productivity gains rather than sharing them with employees, don’t hold your breath for that four-day workweek.