Every Time Machines Threaten Jobs, the Table Gets Bigger

Maor Shlomo cannot write frontend code. He cannot do UI design, doesn’t understand DevOps, has no investor contacts, and lacks enough cash in his bank account to burn for six months.

In early 2025, he opened his computer in his apartment in Israel and fed a product idea to an AI coding assistant. He wanted to build a platform that would allow people who can’t code to quickly build business applications. This idea wasn’t new—several companies were already doing similar things, each with teams of a few dozen people and millions in funding.

He worked alone for three weeks. The AI wrote code for him, designed the interface, ran tests. He was responsible for deciding what the product should look like, what problems it should solve, and what to prioritize.

The product is called Base44. Within the first month of launch, it had paying users. By the fifth month, its monthly profit was $189,000. In the sixth month, Wix made an acquisition offer of approximately $80 million.

One person. Six months. Zero employees. Zero funding.

If you had told this story to anyone in the venture capital world in 2020, they would have thought it was a joke. A person who can’t code, with no team and no money, achieves nearly a million dollars in annualized profit in six months and then gets acquired by a public company for eight figures? The narrative was missing too much: At least one technical co-founder was needed, right? At least one angel round of funding? At least a decent office and three or five engineers?

All these “at leasts” became irrelevant in 2025.

While public discourse is fiercely debating “will AI take your jobs,” a group of people has quietly leveraged AI as a tool to pry open doors previously welded shut by capital, manpower, and technical barriers.

These two things are happening simultaneously, forcing us to rethink a question: Is AI’s impact on employment a story of subtraction or addition?

The answer lies in the last two hundred years of history.

The Loom Broke Two Hundred Years Ago, and the Table Kept Getting Bigger

On a night in March 1811, a stocking factory near Nottingham, England, was smashed. Workers used hammers and axes to destroy the newly arrived knitting machines. They called themselves followers of “General Ludd”—a fictional leader who likely never existed. Over the next two years, similar violent incidents spread to textile factories in Lancashire and Yorkshire. The British government deployed over ten thousand soldiers to suppress the uprisings, more troops than were fighting Napoleon on the Iberian Peninsula at the same time.

The Luddites’ fear was specific: machines wove cloth over a dozen times faster than human hands; weavers would starve.

Were they wrong? In the short term, no. A large number of hand-loom weavers did lose their jobs in those years, wages plummeted, and life was hard. But over the entire 19th century, something strange happened.

Machines crashed the production cost of cotton cloth. Cotton prices fell accordingly. As prices dropped, people who couldn’t afford cotton before could now buy it; those who owned only one shirt could now buy three. Global demand for cotton cloth exploded. Textile mills grew more numerous and larger. Cities like Manchester, Liverpool, and Sheffield saw their populations increase tenfold during the 19th century. The influx of people into cities wasn’t for unemployment—they went to factories, docks, railways, to jobs unimaginable in the countryside.

The “job” of a textile worker was indeed reshaped by machines. But the total employment created by the textile industry chain far exceeded the manual jobs replaced by machines. The total employed population in Britain during the Industrial Revolution didn’t decrease; it doubled.

This pattern reappeared a hundred years later.

In 1899, electricity accounted for less than 5% of the main power sources in U.S. manufacturing. By 1919, it was 50%. By 1929, 75%. Within thirty years, the steam engine was completely replaced by the electric motor.

This time, the object of fear shifted to “electricity.” Large numbers of workers dependent on steam power—miners, boiler operators—felt threatened. The coal industry began a long-term decline.

But electrification brought more than just a power substitution. Electricity allowed factory layouts to shift from bulky structures centered around a central shaft to flexible assembly lines—each machine could have its own motor, placed anywhere. This seemingly minor change completely transformed the organization of manufacturing; Henry Ford’s assembly line was built on the foundation of electrification.

The more profound impact was outside factories. Electricity spawned a whole batch of industries impossible in the steam age: household appliances (refrigerators, washing machines, vacuum cleaners), radio stations, cinemas, elevators (making skyscrapers possible), neon advertising. Each new industry was a new employment pool. In the 1920s in the United States, the urban middle class rose as a social class. Most professions within this class—appliance salespeople, radio program producers, cinema managers, electrical engineers—did not exist in 1890.

Electricity didn’t eliminate jobs. It reduced the cost of “doing things,” so people did more things, and more people were needed to do those new things.

The third panic occurred in the internet age.

In 1995, Newsweek published a famous article predicting that the internet would not have a substantial impact on the economy. This judgment seemed reasonable at the time—what could you do on the internet? Look at web pages? Send emails?

Five years later, the dot-com bubble burst. Numerous .com companies went bankrupt, and hundreds of thousands lost their jobs. “The internet is just a scam” became the dominant narrative.

Another five years later, Facebook launched. Two years after that, the iPhone was released. A few years later, something called the “App Economy” grew from nothing. By 2020, the Apple iOS ecosystem alone directly supported over 2.1 million jobs in the United States. The global app economy was valued at over $6 trillion.

App developers, social media managers, search engine optimizers, Uber drivers, delivery riders, live-stream e-commerce hosts, podcast producers, online therapists, remote freelancers—before 2007, these professions didn’t exist. They didn’t “transform” from old jobs; they emerged entirely from nothing.

By 2023, nearly 64 million Americans were earning money as freelancers or gig workers, accounting for 38% of the entire workforce. The “gig economy” itself was a completely new employment form catalyzed by the internet—a form unimaginable in 1990.

Three technological shocks, three panics, three identical outcomes.

The pattern is clear to the point of being unsettling: New technology emerges → specific jobs are indeed impacted → short-term unemployment for some people → new technology reduces the cost of doing things → more people do more things → new job categories naturally grow from market expansion → within five to ten years, total employment exceeds pre-shock levels.

Some voices will say: AI is different from previous technologies. Previous technologies replaced physical labor or specific cognitive tasks; AI replaces general cognitive abilities. This time, the impact is broader and faster; historical analogies may not apply.

This concern deserves serious consideration. But even the most cautious economists—like 2024 Nobel laureate in economics, MIT’s Daron Acemoglu—have not predicted that AI will lead to long-term mass unemployment. His worry is more about transitional friction: the speed is too fast, reskilling can’t keep up, and certain groups will bear disproportionate costs in the short term. The key variable is not “will technology eliminate jobs” (it will), but “will the newly created jobs outnumber the eliminated ones.”

Historically, every time, the answer has been more.

So the question becomes: What mechanism is driving this pattern? Why does increased efficiency always result in more employment, not less?

Why the Labor Saved Was Never Really “Saved”

“AI will eliminate hundreds of millions of jobs.”

Open any news app, and this judgment can be found within three scrolls. McKinsey says 800 million jobs will be affected globally by 2030. Goldman Sachs says 300 million full-time jobs could be automated. The World Economic Forum says that by 2030, 92 million existing jobs will be replaced by technology.

The volume of panic is enormous, but a set of data is drowned in the headlines.

The same World Economic Forum Future of Jobs Report 2025, on the same page that announces “92 million jobs will be replaced,” also states another number: 170 million new jobs will be created. A net increase of 78 million. This number is equivalent to the entire labor force of Germany.

Most reports only quoted the first half.

This selective blindness keeps happening because people lack a framework to understand “why increased efficiency doesn’t reduce work.” Over 160 years ago, a British economist provided this framework.

In 1865, William Stanley Jevons was staring at coal mine data reports and noticed something strange. The thermal efficiency of steam engines had greatly improved over the past few decades. By logic, the same output should require burning less coal, and national coal consumption should decrease. But the opposite happened: coal consumption surged.

The reason is simple. Efficiency improvements made steam power cheaper, so more industries could afford to use steam engines. Textile mills used them, mines used them, railways used them, ships used them. Each industry expanded capacity, hired more workers, and burned more coal. The cost savings from efficiency were consumed by the influx of new demand, and then some.

Economists later named this mechanism the Jevons Paradox. The rebound effect it describes is playing out exactly as before in the AI age.

AI is making cognitive labor cheaper. Summarizing a legal contract, drafting a set of financial models, translating a technical document, generating a product prototype—tasks that used to take professionals hours can now be completed by AI in minutes.

Intuitive judgment suggests: fewer people will be needed.

But the rebound effect of Jevons Paradox is more honest than intuition. When the cost of doing something plummets, three things happen simultaneously: existing buyers do more (previously doing a financial analysis once a quarter, now doing it weekly); those who couldn’t afford it before enter the market (small business owners who couldn’t afford lawyers can now use AI legal tools to handle contracts); entirely new use cases emerge out of thin air (previously, no one would do sentiment analysis on every customer email, but now it’s possible).

The demand expansion driven by these three channels often far exceeds the tiny reduction from efficiency gains.

Look at the most classic example. In 1969, New York’s Chemical Bank installed the first ATM. Intuitive judgment: a machine can do a teller’s job; the teller profession is doomed. But Boston University economist James Bessen tracked the data over the following decades and found the opposite: ATMs reduced the operating costs of individual bank branches → banks found it cheaper to open new branches → so they opened many new branches → total teller employment actually increased. The tellers’ job content did change—from counting money to customer relations and cross-selling products. Skills upgraded, and wages rose.

The same story happened at supermarket checkout counters (cashiers increased after barcode scanners became widespread), law firms (paralegals increased after electronic discovery software appeared), and the accounting industry. In 1979, when VisiCalc launched, everyone predicted spreadsheets would eliminate accountants. Reality? The U.S. reduced 400,000 accounting clerks but added 600,000 certified accountants—because companies started running more financial models and needed more people to interpret the results (BBC / NPR Planet Money).

The real pattern is: machines reduce the cost of an activity → the total market size of that activity expands → new, more complex human work grows from the expanded market.

The labor saved was never truly “saved.” It was reinvested into things that were previously too expensive, too difficult, or even unimaginable to do.

This mechanism is especially potent with AI. Because AI doesn’t reduce the cost of a specific process in a single industry (like ATMs only reduced the cost of cash handling in banks), but the cost of a general cognitive ability. Writing, coding, analyzing, translating, designing, reasoning—these abilities permeate almost every industry. When their costs plummet simultaneously, the scale of unlocked demand is not just a local expansion in one industry, but a systemic expansion of the entire economy.

So what does this expansion specifically look like? Where do new jobs actually emerge?

Four New Paths Already Growing

At least four paths have already shown clear outlines in reality.

Barrier Shattering. Shlomo’s story is not an isolated case. Around the same time, in Norway, a 20-year-old law school dropout, William Lindholm, used a no-code platform to build a “cake delivery for customer acquisition” platform called Daymaker. He didn’t write a single line of code and was earning $110,000 per month within five months. Austrian developer Peter Steinberger built an AI agent project called OpenClaw in his spare time. It gained 145,000 GitHub stars in 60 days and was acquired by OpenAI three months later.

The commonality of these stories is not “genius entrepreneurship.” The commonality is: AI compressed the product development process that used to require a team of 10 people working for three months into something one person can complete in a few weeks.

Sam Altman said in 2024 that he and a group of tech CEOs were betting on “in what year the first one-person billion-dollar company will appear.” Anthropic’s Dario Amodei gave a probability of 70-80%, with a time window of 2026. In Y Combinator’s Winter 2025 batch, the proportion of solo founders rose from the historical 5-10% to 15-20%. YC partner Jared Friedman’s exact words: “The minimum viable team is shrinking. What used to require 5 engineers now only needs 1 engineer plus AI tools.”

The human barrier to entrepreneurship has been smashed. In the past, building a software product required assembling frontend, backend, design, product management, and someone who understood operations. Just the salaries would burn through hundreds of thousands. This barrier kept 99% of people out. Now AI covers all or part of these roles at the execution level; a person with ideas and business understanding can launch a product. Over 30.4 million Americans are already operating businesses as solopreneurs, with a combined economic output of $1.75 trillion. The expansion of this group with AI tools is the most overlooked growth in today’s job market.

The logic of barrier shattering is very similar to what Shopify did for brick-and-mortar retail. Shopify didn’t “eliminate” retail; it lowered the barrier to “opening a store” from hundreds of thousands to a few hundred, so millions of micro e-commerce businesses that couldn’t have existed before emerged worldwide. What AI is doing to knowledge work is a higher-dimensional version of the same thing.

Entirely New Job Categories. Before 2023, the term “prompt engineer” couldn’t be found in any career guide. In 2024, job postings for this role grew by 135.8%. “Agentic AI” related roles grew even more dramatically, by 985% from 2023 to 2024.

AI trainers, AI ethics auditors, AI safety engineers, AI explainability specialists—these roles hardly existed before 2020 and are now among the fastest-growing categories in the recruitment market. The BLS projects that data scientist jobs will grow by 33.5% from 2024 to 2034, ten times the average growth rate of all occupations.

But direct jobs are just the tip of the iceberg. The larger employment pool is at the indirect level. AI makes personalized education possible, causing roles like “AI-assisted teaching product manager,” “education data analyst,” and “learning path designer” to emerge from nothing. China’s speed in this area is particularly fast. In 2024, the Ministry of Human Resources and Social Security added 19 new job categories at once, including “Generative AI System Application Specialist.” Industry reports predict that China’s AI talent gap will reach 4 million by 2030.

The data annotation industry base in Wuzhong, Ningxia, is an interesting microcosm. In the inland cities of northwest China, young people sit at computers, drawing boxes around images, labeling audio, and classifying text. This is a completely new type of manufacturing catalyzed by the AI industry chain—the manufacturing of data. It doesn’t appear on any list of “jobs AI will eliminate” because this job simply didn’t exist three years ago.

Productivity Multiplier. PwC’s 2025 Global AI Jobs Barometer has a set of data worth savoring: industries with the highest AI exposure (financial services, software publishing) have seen productivity growth soar from 7% to 27% since 2022, nearly a fourfold increase. Per capita revenue growth is three times that of low-exposure industries. With the same number of employees, with AI support, they are producing far more value than before.

If everyone becomes more efficient, shouldn’t fewer people be needed?

Return to the Jevons Paradox. When software development becomes cheaper, companies won’t stop at “doing the same amount of work with fewer people.” They will do more things. Projects previously shelved due to insufficient development resources get started. Products previously considered too niche to be worth making get developed. The total market pie gets bigger, and a bigger pie needs more hands to slice.

NVIDIA’s Jensen Huang revealed a number at GTC 2026: Inside NVIDIA, each human employee corresponds to 100 AI agents, with 75,000 employees paired with 7.5 million agents. His judgment: “In the future, every company’s IT department will be the HR department for AI agents.” In other words: AI agents aren’t replacing people; they are being managed, dispatched, and integrated by people. Managing AI itself is becoming a new type of work that requires a lot of human effort.

Market Unlocking. The final path is the most easily overlooked but may have the most profound impact.

There is a huge amount of demand that people don’t want, but because human labor costs were too high, no one could afford it. Personalized medicine is a prime example. Creating a customized treatment plan for each patient, considering genomic data, medical history, lifestyle habits, drug interactions—in the era of pure human effort, this was exclusive to VIP services at top-tier hospitals. AI has reduced the cost of this analysis by several orders of magnitude. When the cost of personalized medicine drops to a level affordable by ordinary clinics, an entirely new service market is opened up. And this market requires a large number of people—health data analysts, AI diagnostic calibration specialists, personalized plan communicators.

Similar logic is happening simultaneously in education, legal services, mental health, and minority language content. Previously, only English and Chinese users could enjoy high-quality voice assistants. AI’s multilingual capabilities are expanding this service to speakers of Hindi, Bengali, and Swahili. Behind the adaptation for each language is a batch of new jobs for localization testing, content moderation, and cultural adaptation.

Looking at these four paths together—barrier shattering allows more people to start businesses, new job categories emerge from nothing, the productivity multiplier expands the total market, and market unlocking makes previously unaffordable things feasible—they all point to a counter-intuitive but historically well-supported judgment: The jobs created by AI will likely outnumber the ones it eliminates.

But historical patterns and growth paths are predictions. It’s worth looking at what the data from 2024-2025 actually shows.

The Volume of Panic and the Data of Reality Don’t Match

First, look at the ammunition on the panic side.

In 2025, approximately 55,000 jobs in the U.S. were directly attributed to AI-related layoffs. The Stanford Digital Economy Lab tracked that junior hiring at the 15 largest tech companies dropped by 25% from 2023 to 2024. Employment of 22- to 25-year-olds in AI-exposed occupations fell by 13% from the end of 2022 to mid-2025. These numbers are real, and behind them are real people experiencing real difficulties.

But they represent a tiny fraction of overall employment data. AI-induced layoffs accounted for only 4.5% of all U.S. unemployment in 2025. Goldman Sachs’ estimate is even more conservative: even if all current AI use cases were fully deployed, only 2.5% of U.S. jobs would face replacement risk.

Now look at the other side.

The number of AI-related jobs is growing at a completely different rate. In Q1 2025, the U.S. had 35,445 open AI-related positions, a 25.2% year-over-year increase. AI engineer job postings grew by 143.2%. Prompt engineer postings grew by 135.8%. The frequency of mentions of “AI” in job postings grew by 56.1% year-over-year in 2025, and this was on top of 120.6% growth in 2024.

PwC’s analysis, covering nearly 1 billion job advertisements across six continents, provided a key finding: in industries with the highest AI exposure, even those “most easily automatable” jobs saw employment growth. The growth rate was slightly slower than in low-exposure industries (38% vs. 65%), but the direction was the same: all rising.

If AI were truly eliminating jobs on a massive scale, employment data in high-exposure industries wouldn’t look like this.

Wage data offers another perspective. The wage premium for AI skill positions doubled from 25% to 56% within a year. Wage growth in high AI-exposure industries is twice that of low-exposure industries. In the UK, the wage premium for possessing AI skills (23%) even exceeds the premium for holding a master’s degree (13%). If a technology were truly devaluing workers, it wouldn’t simultaneously make workers who master that technology more expensive.

Of course, there is a dark side to this data that cannot be ignored. What’s growing are jobs requiring AI skills; what’s shrinking are jobs not requiring AI skills. During the same period that AI skill job postings grew 7.5% year-over-year, total job postings across all occupations fell 11.3% year-over-year. The window for skill migration presents real pressure: if you don’t learn new skills, the job market you face is indeed shrinking.

This is perfectly consistent with the pattern of every historical technological shock. The Luddites’ pain wasn’t fake, but the long-term expansion of the textile industry wasn’t fake either. Coal miners in the electrification era did lose their jobs, but appliance factories needed more workers. The key variable has always been the same: how long is the transition period, and can the speed of retraining keep up with the speed of replacement?

WEF data provides a specific time frame: by 2030, a net increase of 78 million jobs. But this 78 million is predicated on the fact that 86% of employers’ required skills are changing rapidly—in AI-exposed jobs, the speed of skill changes demanded by employers is 66% faster than a year ago.

The picture painted by the data is not “AI eliminates work,” nor is it “all is well.” It paints a labor market in high-speed reorganization: old jobs are contracting, new jobs are expanding, the expansion speed is currently faster than the contraction speed, but the skills gap in between is real and urgent.

The panic narrative captures the contracting half. The optimistic narrative captures the expanding half. Both are true, but the net effect in the data is positive.

What You Should Prepare Is Not a Resume, but a Toolbox

Back to Maor Shlomo.

One person, six months, from zero to an $80 million exit. He’s not a genius programmer, nor a second-generation founder with a silver spoon. He was just half a step ahead of others in understanding one thing: AI has reduced the cost of “making a product” to near zero, but the ability to “discover a problem worth solving” remains scarce.

What he did was find a specific pain point (non-technical people wanting to quickly build applications), then used AI as leverage to move a market that previously required a team to move, with the scale of one person.

This story contains a signal more useful to most people.

The employment logic in the AI era is undergoing a structural shift. In the past, your value was mainly determined by “what you could do”—code, design, do financial analysis. These are skills that can be trained, replicated, and automated. When AI can produce usable code, design drafts, or financial models in minutes, the premium for pure “execution skills” will be compressed.

But what AI is not good at is becoming more valuable: judging which problems are worth solving, making trade-offs in ambiguous information, integrating AI’s output into a scenario with real-world friction, making decisions under uncertainty and bearing the consequences.

A detail easily overlooked in PwC’s report: in AI-related hiring, design capabilities have surpassed pure technical skills to become the most needed skill by employers. “Design” here doesn’t mean drawing; it’s the ability to systematically understand user needs, define problem boundaries, and plan solutions. The Autodesk 2025 AI Employment Report states verbatim: human-centered thinking is in rapidly increasing demand in AI development.

The stronger AI gets, the scarcer the ability to “understand people” becomes.

Facing this change, anxiety is natural. But more useful actions after anxiety are three things.

Treat AI as a capacity amplifier. In your current work, there is certainly a lot of repetitive cognitive labor—writing reports, data cleaning, organizing meeting minutes, translating documents, drawing prototypes. Hand these over to AI. The time saved is not for resting; it’s for doing things AI cannot do: talking face-to-face with clients about needs, making judgments among three conflicting proposals, coordinating a cross-departmental issue that no one owns but everyone is affected by.

Learn to drive AI. The biggest trap is spending a lot of time learning what AI can already do. A more leveraged investment is learning how to command AI: how to break down a complex task so AI can handle it, how to judge where AI’s output is trustworthy and where it’s not, how to orchestrate multiple AI tools into a workflow that runs. Jensen Huang said that in the future, every company’s IT department will be the HR department for AI agents. Managing AI itself is a new, high-value type of work.

Go after opportunities where AI has lowered the barrier. Previously wanted to make a side project product but gave up because you couldn’t code? Previously wanted to create an online course in a minority language but shelved it because production costs were too high? Previously wanted to build a vertical tool for your industry but never started because it required a team? AI is dismantling these barriers one by one. The 30.4 million American solopreneurs and the AI one-person businesses emerging across China are footnotes to this trend.

Two hundred years of data and current reality point to the same judgment: AI will eliminate some jobs, then create more. The net effect is positive. But this “positive” won’t automatically fall on everyone’s head. It falls on those willing to pick up new tools, learn new skills, and enter new markets.

Shlomo didn’t wait until everyone agreed that AI entrepreneurship was feasible before he started. While others were still debating “will AI replace programmers,” he had already used AI to finish the product, launch it, collect the money, and receive an eight-figure check.

What he picked up wasn’t a better resume; it was a bigger toolbox.

Appendix: Sources Cited in This Article

WEF Future of Jobs Report 2025: 170M new jobs / 92M disappeared / net increase of 78M https://www.weforum.org/press/2025/01/future-of-jobs-report-2025-78-million-new-job-opportunities-by-2030-but-urgent-upskilling-needed-to-prepare-workforces/
PwC 2025 Global AI Jobs Barometer: 56% wage premium / 4x productivity growth / 38% employment growth https://www.pwc.com/gx/en/news-room/press-releases/2025/ai-linked-to-a-fourfold-increase-in-productivity-growth.html
AI job growth data (Veritone / Autodesk / BLS): https://novoresume.com/career-blog/ai-job-creation-statistics
ATM and bank tellers (James Bessen / AEI): https://www.aei.org/economics/what-atms-bank-tellers-rise-robots-and-jobs/
VisiCalc and accountants (BBC / NPR Planet Money): https://www.bbc.com/news/business-47802280
Jevons Paradox and AI (MindStudio): https://www.mindstudio.ai/blog/jevons-paradox-ai-human-work-demand
One-person companies and solo founders (Forbes / Taskade): https://www.forbes.com/sites/sandycarter/2026/04/04/openai-called-the-one-person-ai-startup-and-three-founders–proved-it/
British textile industry and Industrial Revolution: https://www.worldhistory.org/article/2183/the-textile-industry-in-the-british-industrial-rev/
American electrification history (Smithsonian): https://americanhistory.si.edu/explore/exhibitions/american-enterprise/online/corporate-era/generating-change
iOS App Economy employment (Apple Newsroom): https://www.apple.com/newsroom/2020/09/ios-app-economy-creates-300000-new-us-jobs-as-developers-adapt-during-pandemic/
Gig Economy (Investopedia): https://www.investopedia.com/terms/g/gig-economy.asp
China AI employment and talent gap (Global Times): https://www.globaltimes.cn/page/202502/1328588.shtml
U.S. Solopreneur data (CNBC): https://www.cnbc.com/2025/09/22/how-to-start-business-ideas-income-opportunities.html