Over the weekend, I went digging for evidence that AI can, will, or has replaced a large percent of jobs. It doesn’t exist. Worse than that, actually, there’s hundreds of years of evidence and sophisticated analyses from hundreds of sources showing the opposite is true: AI will almost certainly create more jobs than it displaces, just like thousands of remarkable technologies before it.

I don’t want anyone to think I’m raining on this parade without first attempting to convince myself that the opposite was true, and that AI really would be the first technology in 120 years to displace a massive portion of the workforce. So, dry though it may be, let’s walk through the logic together.
The majority of statements that have received press (and there have been dozens in the last 5 years) center on the claim that AI will destroy 20-50% of the current need for human labor. I’ll attempt to address each of the most robust points inherent in those arguments, rather than trying to argue that any one innovation or upgrade to a model’s capability hasn’t done it yet (or won’t):
- If AI is going to make such a huge percentage of jobs redundant, there must be historical analogies–i.e. other technologies that massively upended labor markets. What are these and how have they affected jobs in the past?
- MIT’s Technology Review noted that this “fear of new tech taking jobs,” is far from new. The automation of farm work is the most notable and most labor-impacting example we have from history, rapidly unemploying a huge portion of human beings in the developing economies of the late 19th and 20th centuries. And yet, at the conclusion of this era (~1940s/50s), the conclusion was that “technological unemployment is a myth,” because “technology has created so many new industries” and has expanded the market by “lowering the cost of production to make a price within reach of large masses of purchasers.” In short, technological advances had created more jobs overall.
- Last year, Quarterly Journal of Economics published a groundbreaking study on how technological innovations have impacted labor forces across industries since 1980. MIT did a nice summarization: “the number of studies that support the labour replacement effect is more than offset by the number of studies that support the labour-creating/reinstating and real income effects.”
- This 2023 paper looked at 127 previous studies of technology supposedly replacing labor forces from the 18th century to the present, concluding that “the labor displacing effect of technology appears to be more than offset by compensating mechanisms that create or reinstate labor.”
- The Economic Policy Institute did a deep dive into what drives labor market demand and unemployment, concluding: “Productivity growth (which technology sometimes enables and other times drives) has not historically been associated with higher unemployment or higher inequality,” and that “Anxieties over widespread technology-driven unemployment lack an empirical base.”
- Perhaps the closest analogy to AI is the personal computer revolution of the 1980s. Millions of jobs in communication, documentation, research, analysis, and engineering became obsolete within a decade, and yet, the McKinsey Global Institute concluded in 2018 that “We tallied up all the jobs destroyed in the US since 1980 as a result of the rise of personal computing and the Internet, and it’s about 3.5 million,” but “When we add up all the jobs created, we find that over 19 million jobs have been created as a result of the personal computer and Internet. We see a net gain of 15.8 million jobs in the US over the last few decades. That’s about 10 percent of the civilian labor force today.”
- If AI is going to have these massive impacts but hasn’t yet, why not?
- Folks who claim AI will destroy the labor market have claimed this radical change is “only a few years away,” “on the immediate horizon,” or “imminent,” for the last 5 years, yet we’re at historically low unemployment (yes, even accounting for underemployment and the way the BLS counts employment). The US labor market is within a single percentage point of its post-war unemployment low, measured in 1953 at 3.4%.

- If AI is killing jobs, it’s doing so at an imperceptibly slow rate; why could that be? Is it still too early? Did other technologies take a long time to show their impacts on labor markets?
- The broad consensus from rational industry observers, analysts, economists, and even AI-hyped technologists is that the end of cheap money (i.e. higher US interest rates) has driven most of the lower-than-pre-pandemic-demand for entry-level talent (just as it has in times of inflation-fighting interest hikes of the past).
- Machine-learning, the technology underpinning AI, has been around for decades, with widespread adoption in tech companies between 2006-2013. The current generative-AI era, based on the transformer architecture model, kicked off in 2017, with significant public examples and tech adoption from 2018-2020. Most of the current, press-driven AI hype cycle, however, skyrocketed in late 2021 with OpenAI’s release of GPT-3 (longtime readers here will recall that Britney Muller showed off techniques extremely similar to what’s now associated with modern LLMs back in July 2018).
- We’ve had 15-20 years of robust machine learning development and adoption, and another 5-10 years of broad LLM/generative AI adoption, improvement, and usage, yet labor market fluctuation has been far more dependent on other factors: the Covid pandemic itself, the post-pandemic surge and decline in tech hiring, inflation-fighting tactics by government banks, and (most recently), a renewal of early-20th-century-style tariffs and trade wars. When controlling for these events.
- The effects of previous technological advancements also took time, but the most salient examples (of farm equipment in the 1910-1920 era and the personal computer in the 1980s) showed millions of displaced workers within 5 years. AI’s slower changes bode poorly for the argument that it will have a larger impact than those events.
- Even if one assumes that AI was the only contributor to labor market changes between 2021-2025, the change has been incredibly slight, *even* in the software engineering market where it supposedly has the greatest impact. There was a greater loss (nearly 150%) in percentage of software engineering jobs between 2019-2021 than from 2021-2025.
- I found it particularly revealing that one of the most commonly cited examples of AI killing labor needs in the software field is the death of StackOverflow, and yet, a robust analysis of that site’s usage from 2008-2020 shows that “What really happened is a parable of human community and experiments in self-governance gone bizarrely wrong.”
- However, it seems likely that the perception of AI and its adoption are slowing hiring in the software engineering market in the post-bubble-popping era (2024-25). This thoughtful analysis by Gergely Orosz concludes a well-visualized, data-driven walkthrough with: “LLMs are a leading cause of the fall in software developer job postings: there’s uncertainty at large companies about whether to hire as fast as previously, given the productivity hype around AI tooling, and businesses are opting to “wait and see” by slowing down recruitment, as a result.”
- It strains credibility to look at the data, history, and analyses and conclude that AI will eventually kill 20-50% of all jobs, when its largest impact in the prior 5-20 years of adoption (depending on one’s starting point) is ~10% variation in a job sector that employs ~1% of US workers.
- Assuming AI will have an effect similar to 20th Century farm equipment’s on agriculture, why will that labor force behave differently to their 20th Century counterparts (and either refuse to or be prevented from finding new jobs)?
- This point is hard to find citations for, given that it’s a future-looking, theoretical assertion. We can, however, compare the impact of the tractor (and farm machinery more broadly) on the economy from 1910-1960.
- Tractors and farm equipment resulted in the shutdown of a huge number of farms, and a decline in the number of people employed in farming, from ~33% to ~2% of the labor force (notably, even that massive upheaval was less significant than the prognostications by tech company leaders that AI will displace half of all jobs). Nothing like it has happened in the American economy since, and only the industrial revolution of the 18th/19th centuries can compete in scale of transformation.
- A superb breakdown of farm machinery’s impact on a sector that employed more than a quarter of all Americans comes from Olmstead and Rhode at UC Davis:

- Is it possible that AI will do to broad sectors of the economy what mechanized farm equipment did to agriculture?
- Rationally, it’s difficult to fathom generative AI having a greater economic and labor-force impact than the PC revolution of the 1980s. AI makes many tasks more efficient, but evidence that it can wholesale replace entire human functions in a tractor-like way is pure speculation that exists in imagination, not reality.
- The core assertion by the “AI will replace 20-50% of all jobs” crowd seems to be that the past 20 years of machine learning and generative AI improvements are not indicative of what will happen in the future: a leap in capability that will enable company management to instruct an AI on a job function (“get us press,” or “optimize our marketing campaign,” or “record and audit our financials” ) and rely on machines to correctly determine what needs to be done, how to do it, and then complete all associated tasks with little to no human supervision, intervention, or additional labor.
- It’s impossible to argue against the assertion that AI will do what’s described above, because it’s based not on objective data, but rather on subjective belief about a possible future. Fighting about what someone believes may come about in the future is generally non-productive, so I’ll avoid that to spare us all a lot of wasted time 😉
I’ll move on from the dry argument analysis and citation process and attempt to summarize (and opine on) what’s really going on here.
Leaders of AI companies, and some AI proponents, marketers, journalists, and even critics have found that when they make scary predictions about their field destroying the job market, press and media eat it up. This media coverage, because it’s scary and the AI hype cycle is in full swing, draws clicks. Those clicks lead to employees, managers, and leaders at other businesses being scared into learning and adopting AI in their businesses.
Incentive also exists for those who criticize AI, AI companies, or their ethics/models/practices: these folks also benefit directly from the attention they earn when they amplify the message of AI as a job destroying technology.
If you’re feeling like the “AI will take all our jobs” discussion is familiar, you’re in good company. Many others have pointed out the similarities to stories like:
Source: Jalopnik

Source: InsideHook

Source: Honest Jobs
Mechanization really did take jobs from farm workers. Automation took jobs from manual laborers. The PC took jobs from clerical and communication workers. But, all of these resulted in greater productivity, employment, and more optionality for workers. It’s both anti-historic and anti-evidence that AI will somehow prove to be the exception.
Could AI, along with thousands of other impactful technological, political, social, demographic, and black-swan-event changes permanently alter the employment landscape in our lifetimes? Absolutely. In fact, one of my favorite stats from this overly-ambitious weekend of research was MIT’s estimation that 60% of employment in 2018 was in types of jobs that didn’t exist before 1940.
By the time I’m in my 80s, y’all better have destroyed more than half of all the existing jobs, and that’s just to keep up with the 20th Century’s pace of change. But, don’t expect AI to do it for you in the next decade; that’s just marketing.
p.s. If you’re looking for the TL;DR, Ed Zitron on Bluesky has got you:
p.p.s. I agree there’s evidence that this fear-based marketing campaign has been successful enough to disrupt some hiring, especially for early-stage jobs in a few tech-heavy fields. But squinting at the evidence, it’s <0.1% of jobs (<200,000 total) being affected, and even here, the unbalanced capital vs. labor market is a far more compelling explanation.