In early 2026, something shifted in the conversation about AI and jobs. It stopped being theoretical. A string of high-profile studies and reports landed within weeks of each other, each painting a different but overlapping picture of what AI is doing to the labor market right now, and what it might do next. Some offered data. Others offered scenarios. One crashed the stock market. Here's what each of them found, and what it all means when you put the pieces together.

The Anthropic study: AI is barely getting started, but the cracks are showing

On March 5, 2026, Anthropic published "Labor market impacts of AI: A new measure and early evidence," introducing a metric called observed exposure. The idea is straightforward: instead of just asking what AI could theoretically do, measure what it's actually doing in professional settings by analyzing real-world usage data from Claude, Anthropic's AI model. The gap between theory and reality turns out to be enormous. For computer and math occupations, large language models could theoretically handle 94% of tasks. In practice, Claude currently covers just 33%. Office and administrative roles show a similar pattern, with 90% theoretical capability but a fraction of that in actual use. The most exposed occupations today are computer programmers (75% task coverage), customer service representatives, and data entry keyers (67%). At the other end, 30% of workers have zero AI exposure, including cooks, bartenders, mechanics, and dishwashers, jobs requiring physical presence that no language model can replicate. The researchers found no systematic increase in unemployment for workers in highly exposed occupations since late 2022. But they did find something concerning: a 14% drop in the job finding rate for workers aged 22 to 25 in AI-exposed fields, compared to 2022 levels. The problem isn't layoffs. It's that young people aren't getting hired into these roles in the first place. The demographic profile of exposed workers also challenges assumptions. The most at-risk group is 16 percentage points more likely to be female, earns 47% more on average, and is nearly four times as likely to hold a graduate degree compared to unexposed workers. This is the financial analyst, the software developer, the lawyer, not the warehouse worker. The researchers explicitly named the scenario that keeps economists up at night: a "Great Recession for white-collar workers," in which unemployment in exposed occupations doubles from 3% to 6%. It hasn't happened yet. But their framework shows it absolutely could, and they would be able to detect it.

The Citrini report: the scenario that crashed the market

On February 23, 2026, Citrini Research, an independent research publication founded by James van Geelen, published "The 2028 Global Intelligence Crisis," a speculative scenario co-authored with Alap Shah of Lotus Technology Management. Written as if it were a memo from June 2028, it described a world in which AI had already triggered an economic disaster. The scenario goes like this: AI capabilities improve rapidly, making it vastly easier to build software, automate customer service, and handle professional tasks that once required teams of white-collar workers. Companies respond by cutting headcount to protect margins, which triggers a negative feedback loop. White-collar workers, who represent roughly 50% of US employment and drive about 75% of discretionary consumer spending, lose their jobs or see their earning power collapse. Consumer spending craters. The S&P 500 falls 38%. Unemployment surges to 10.2%. The report also introduced the concept of "agentic commerce," where AI agents optimize away the human inefficiencies that many businesses depend on for revenue. Insurance customers get automatically switched to cheaper plans. Real estate transactions bypass agents. Payment processing gets replaced by cheaper alternatives. Each optimization eliminates revenue that once paid for human labor. Citrini was careful to label this a scenario, not a prediction. But the piece went viral, racking up 28 million views on X and contributing to a significant sell-off in software and tech stocks on February 23 and 24. It prompted Citadel Securities, a firm not known for publishing market commentary, to issue a rebuttal arguing that fundamentals were being overlooked. Evercore analysts called the piece "thought-provoking" but "implausible," arguing it assumed the government would do nothing in response. Economist Noah Smith dismissed it as "a scary bedtime story." But the market reaction showed that the scenario had touched a nerve.

The HBR survey: layoffs driven by anticipation, not performance

In January 2026, Thomas H. Davenport and Laks Srinivasan published findings in Harvard Business Review from a survey of 1,006 global executives conducted in December 2025. The central finding was striking: AI is behind at least some layoffs, but these are almost completely driven by anticipation of AI's impact, not its actual performance. In other words, companies are cutting jobs and slowing hiring based on what they believe AI will eventually be able to do, even though the technology hasn't yet delivered on those promises. This creates a paradox: the job losses and slowed hiring are real, even as companies are still waiting for generative AI to prove itself. This helps explain the strange dynamics in the current labor market, where hiring has cooled without a corresponding spike in unemployment. Companies are hedging against a future they expect but haven't yet arrived at.

Gartner's forecast: not an apocalypse, but chaos

Gartner's 2025 AI Job Impacts Analysis, published in November 2025, offered perhaps the most nuanced framing. Its core prediction: starting in 2028 to 2029, AI will create more jobs than it eliminates. But it will also fundamentally transform over 32 million roles per year, roughly 150,000 jobs changing every day. "AI will not trigger a job apocalypse, but job chaos," said Gartner analyst Helen Poitevin. The distinction matters. Gartner's position is that AI's impact on global jobs will be roughly neutral through 2026, with the real disruption beginning in 2028 when the rate of job redefinition accelerates. The challenge isn't mass unemployment but mass reskilling: millions of workers will need to learn new skills or move into redesigned roles as AI gets embedded into business and IT processes. The implication is that companies and governments that prepare for this transition, by investing in reskilling and redesigning work rather than simply layering AI on top of old structures, will fare far better than those caught off guard.

Goldman Sachs: the moderate case

Goldman Sachs Research, in an August 2025 analysis, offered a comparatively optimistic view. The firm estimated that unemployment would increase by only half a percentage point during the AI transition period as displaced workers seek new positions. If current AI use cases were expanded across the economy and reduced employment proportionally to efficiency gains, an estimated 2.5% of US employment would be at risk of related job loss. The occupations Goldman identified as most at risk, computer programmers, accountants, auditors, legal assistants, administrative assistants, and customer service representatives, align closely with the occupations flagged by Anthropic's observed exposure metric. Goldman's core argument is that while AI will undoubtedly displace workers, historical precedent suggests the economy adapts. Firms that use AI extensively tend to be larger, more productive, and pay higher wages. They also grow faster: a large increase in AI use is linked to about 6% higher employment growth and 9.5% more sales growth over five years.

What the studies agree on

Despite different methodologies and levels of alarm, these reports converge on several points: White-collar workers are the most exposed. Unlike previous waves of automation that hit manufacturing and routine blue-collar work, AI is coming for cognitive, non-routine tasks. Programmers, financial analysts, customer service workers, and administrative professionals are at the front of the line. Young workers are being hit first. Multiple studies found that the earliest signal of disruption isn't layoffs but slowed hiring, particularly for workers aged 22 to 25. The Anthropic study found a 14% drop in job finding rates for young workers in exposed occupations. A separate study by Brynjolfsson et al. found a 6% to 16% fall in employment in AI-exposed roles among workers in the same age bracket. The gap between capability and adoption is still wide. AI can theoretically perform a large share of white-collar tasks, but actual usage remains a fraction of what's possible. Legal constraints, technical limitations, and the need for human oversight are slowing things down. This is a window, not a permanent barrier. Anticipation is already causing real effects. The HBR survey showed that companies are making hiring and staffing decisions based on AI's expected trajectory, not its current performance. This means the labor market impact is partly self-fulfilling. The macro risk is real but uncertain. The Citrini scenario may be speculative, but the underlying logic, that white-collar spending drives the US economy and white-collar jobs are disproportionately exposed, is acknowledged across multiple reports. Whether this leads to a consumption-driven recession depends on the speed of disruption and the policy response.

What to watch for

The next 12 to 18 months will be critical. Here's what will clarify which scenario we're heading toward:

The studies disagree on how bad things will get, but they agree that something significant is underway. The question is no longer whether AI will reshape the job market. It's how fast, how deep, and whether we'll be ready.

References