The AI Labor Shock: Early Evidence of How Algorithms Are Replacing the Middle Class

A 30-year-old woman named Jade works as an insurance tech in Raleigh, North Carolina. She spends part of her day optimizing AI workflows. The labor is lucrative. The work is authentic. Additionally, she claims that at times, the work feels like constructing the framework for her own replacement. “Half the time it feels like the whole job is just AIs talking to each other, with barely a human involved,” she said to The New Republic. “It doesn’t feel good to build a flow with my labor that could potentially replace me — not just another person, but me.”

This is the psychological front of a structural economic change that the data is just now starting to reveal. The timeline is still genuinely uncertain, the forecasts vary by orders of magnitude, and the academic evidence is inconsistent and disputed. Even though they defy easy summarization, the preliminary figures from labor market studies are startling enough to be taken seriously.

A 2025 study that was published in the Journal of Economic Behavior & Organization looked at over three million job postings on a significant online freelance marketplace both before and after ChatGPT launched in November 2022. After grouping those postings into 116 different skill clusters using topic modeling, the researchers categorized each cluster as either unaffected, complementary to AI, or substitutable by AI. The outcomes for the categories that could be substituted were striking.

Compared to the trend, writing for real estate listings decreased by 52%. Copywriting on “About Us” pages for businesses decreased by 59%. Western European languages saw a 23% decline in translation. Demand for short-term contracts fell by 20 to 50% across all substitutable clusters. In the clusters where people complement the tools rather than compete with them, the freelance market as a whole expanded rather than contracted. Jobs in machine learning increased by 24%. The development of AI chatbots increased by 179%. A notable divergence was concealed beneath the overall image.

Peripheral skills are not the substitutable categories that declined the most. Writers, translators, junior programmers, and content creators are the types of jobs that have sustained middle-class independent contractors for many years. Celina Odeh, a 30-year-old commercial photographer in Greenville, South Carolina, compared her days on set to a parlor game of elimination: which skills would AI eliminate next? To keep herself sane, she has taken up knitting, which is a particular and strangely poignant detail. It accurately depicts the specific type of anxiety this causes, which is chronic, pervasive uncertainty with no obvious course of action rather than fear of an impending disaster.

Without resolving the timeline, the macroeconomic estimates increase the scale. According to a McKinsey study from 2023, generative AI could automate between 60 and 70 percent of worker tasks in specific job categories. Roughly 80% of US workers could see at least 10% of their tasks impacted by large language models, with roughly 19% seeing more than half of their tasks affected, according to researchers connected to OpenAI. According to MIT researchers, 12% of the US labor market could be automated at a reasonable cost today, and that number could rise as capabilities advance. The consistent direction of the estimates is important, even though the ranges differ because they measure different things, such as potential exposure rather than actual displacement.

Several economists have pointed out that the lack of a clear skill prescription makes this specific technological shift more difficult to manage than others. “There’s no obvious solution like ‘Get a college degree’ or ‘Get a STEM degree,'” stated Victor Chen of Virginia Commonwealth University. Planning your career, let alone your life and your children’s futures, becomes challenging as a result. There was at least a tenable narrative about retraining for cognitive or service roles during earlier automation waves that hollowed out manufacturing. The current wave specifically targets cognitive work, especially the middle-tier cognitive work that was meant to be replaced, such as competent coding, standard legal drafting, competent writing, and routine financial analysis.

In a 2024 paper, David Autor of MIT, who has spent decades researching how technology changes labor demand, made the case that, if implemented consciously by businesses and governments, artificial intelligence (AI) could actually bring back middle-class, middle-skilled jobs rather than eliminate them. The claim is that when AI tools are paired with employees who actually possess domain knowledge and judgment, those employees may become more productive and, consequently, more marketable. He makes it clear that this is a case for what is feasible under specific policy circumstances rather than a forecast. That is not the case with the current deployment pattern, which places the majority of the burden on the most automatable tasks and reaps the majority of the benefits for the companies that automate.

When asked about the shift, Nobel laureate Joseph Stiglitz expressed more pessimism. “Looking back, we didn’t manage industrialization well, and we don’t appear to be managing this well, either,” he stated. Because publicly traded companies’ incentive structures favor cost reduction over worker investment, he worries that the market won’t self-correct toward the productive outcome Autor describes. The Academic-Industry Research Network’s William Lazonick put it more succinctly: “Companies want to use AI as a way to reduce costs, increase profits, pay massive dividends, and do stock buybacks.” For human workers, what does any of that accomplish?

Lexi, a 56-year-old brand copywriter in Brooklyn, has spent 30 years honing the specialized skill of assisting companies in determining what to say and how to say it. These days, she observes AI creating versions of that work in a matter of minutes, though she characterizes them as having “near-zero personality.” She is spending down savings and hasn’t found a clear next chapter. “I’m not sure who’s reading that, either,” she says of the writing she still produces. That sentence contains a lot: the devaluation of craft, the erasure of audience, the looping uncertainty of what any of this is actually for now.

Observing these individual testimonies and preliminary data suggests that this is more of a gradual reorganization with winners and losers dispersed unevenly across industries, age groups, and skill levels than a singular crisis. According to the freelance data, the change is already occurring in some categories at a discernible rate. Whether it compounds further into the broader labor market, or whether policy and corporate strategy find a way to redirect it toward the more productive outcome According to Autor, this will likely be the most important economic issue of the upcoming ten years. The information is preliminary. You can see the trend.