26:29
The third video we ever made on this channel was about automation. This was way before AI as we know it today, before LLMs and ChatGPT made a big splash, and before AI tools became something that could write a legal brief or pass a medical exam on a Tuesday afternoon. Back then, when we talked about automation, we were referring to robots on factory floors and software making rows of accountants redundant. The future felt like a question of if and when rather than now and how fast. So, we did what economists do, and we ran a thought experiment. What would a fully automated world actually look like? That was July 2019. In November
2022, ChatGPT launched, and almost overnight, the automation debate shifted from something economists worried about in academic papers to something people were experiencing on their lunch break. Now, a lot of good economics is really just looking back at the theories you put forward and asking how well they held up to reality. That's exactly what this compilation is. We've put together three videos from this channel spanning seven years. That original thought experiment from 2019 with three possible futures for a world where machines do most of the work, a reckoning with what AI is already doing to real economies from 2025, and an MIT study from earlier this year that changed how we measure what this
technology is actually doing to us. Together, they trace how our understanding of this has evolved and how much we're still only beginning to figure out. So, how has the automation story changed since we first covered it? Which of our predictions held up and which ones didn't? And where does all of this leave workers and economies today? Amidst all of this AI change, data has never been worth more as it is used to train and develop AI models. While that's great for making AI better, it also means your data has never been more insecure. That's why I use Surfshark. It's a VPN, a virtual private network, that encrypts your internet traffic and allows you to access over 4,500 servers in 100 countries.
It protects all your personal data and does so across unlimited devices, but it can do much more than just keep your information safe. Surfshark allows you to swap your digital location, letting you look for deals around the world and access content outside your region. While security is great, what sets Surfshark apart is their CleanWeb feature. It blocks all pop-ups so you can stay focused on what you're doing. Surfshark is offering Economic Explained viewers a special deal. With our code, you can get four extra months for free. Go to surfshark.com/economics or use code economics at checkout to get four extra months of Surfshark VPN.
Automation is inevitable and it is coming for you. Steel made bronze obsolete. The internal combustion engine made steam power obsolete. The self-checkout at your local grocer has made teams of workers obsolete. Machines and automation are coming to make you obsolete. If you don't believe me, then that is fine. At the end of the day, I am a stranger who says he is an economist on the internet and any of those three things should make you wary of a naysayer on their soapbox screaming that the end is nigh. But, it really might be this time. Automation for the most part is already here and it is silently and slowly pushing the working population to the wayside. And this gentle pressure all
has to do with good old-fashioned supply and demand. This is a demand and supply graph. This is used by economists to illustrate the price of things given a certain level of, well, supply and demand. Let's say this is the demand and supply for apples on this graph. The x-axis represents the quantity of apples and the y-axis represents the price of apples. Beyond that, there are two important lines. One for consumers, which makes up the demand line. This line slopes downwards because if apples were selling for two cents each over here, pretty much everyone would want to buy as many apples as they could because, well, they're just so cheap. But, over here, if apples were $10 each,
not nearly as many people would want them. Sure, there might still be apple connoisseurs that live and die by the apple, but most people would rather just go and buy a banana. I mean, at the end of the day, what can a banana cost? The same is true, but in reverse for this line, the supply line. If apples were 2 cents each, nobody would bother growing them. It would cost far more in farming and shipping than a farmer could ever hope to make as profit off these 2 cent apples. So, they would just go and grow something else. Likewise, if apples got to a point where they were selling for $10 each, everybody would want in on this apple growing business.
Corn farmers would plow their fields to make rooms for apple orchards, and the supply of apples would skyrocket this point. But, remember, nearly nobody wants to buy a $10 apple. So, in reality, this point here, where the lines cross and supply meets demand perfectly, is where apples would normally trade. A reasonable amount of apples would sell for a reasonable price, and everybody would feel like they are getting a fair deal. This whole apple business is the foundation of transactional economics. But, what a lot of people don't realize is the same goes for their jobs. So, what happens if we switch out apples for accountants? And you may say, "Well, don't be silly. You can't buy an
accountant." And yes, slavery is more or less frowned upon these days, but what you can do is pay for an accountant in the form of a salary. Just the same way, if an accountant got paid $10 a year, every business in the world would want to hire accountants. They are so cheap to employ that even if they only just got the morning coffee, it's worthwhile having them around. But, in this scenario, not many people would ever want to work as accountants. But, likewise, if they were to earn $1 million a year, almost everybody would want to be an accountant. People would switch from other professions, and
universities training accountants would fill up fast. Unfortunately for these dreamy bean counters, companies only need to hire so many accountants, and so most of them would not really be able to find work. Again, where these two lines cross is the reality in the market today, where about 1.26 million accountants work in the United States for an average salary of about $60,000. This all makes sense so far, but where it gets interesting is where these supply lines move. If an office opens up in the Philippines, for example, that allows companies to outsource their accounting work overseas, what that is
doing is effectively increasing the supply of accountants. This means that any given salary there would be more accountants willing to work. On the more sinister side, though, it means that companies will be able to employ the same amount of accountants for less money. This is one of the leading causes of wage stagnation in This is affecting many, many wealthy economies today. The same thing can happen here on the demand side, though, in this equation. A company will see that five accountants with electronic calculators can do as much work as 20 accountants with abacuses, and two accountants with Excel
and accounting software will be able to do as much work as five accountants with calculators. As capital assets in the form of technology improve, the same amount of work can be done with less and less people. This means that a company that used to employ 200 accountants will now only need to employ 20 accountants, and those skilled accountants will be forced to compete on who is willing to work for the lowest of wages. Abacuses to calculators to desktop computers, if we take this to its logical end, though, we find that a highly automated machine that can do the work of those accountants without any accountants at all might be the logical end. So, what will
a world look like when almost every job is automated this way? Predicting the future is impossible, unless you are The Simpsons. Predictions about what computers will mean for the future have been made for as long as computers have existed, and for the most part, these predictions have been hilariously bad. So, I am going to try and avoid this prediction being based off what automation will look like and focus more on what the whole economy will look like. A fair warning is that economists are not much better than computer scientists when it comes to predicting the future. But, most reputable economists that have been brave enough to make predictions have stated the machine future is likely to look like one of three possible
outcomes: the good, the bad, and the ugly. I like my dessert first, so let's take a look at the best possible outcome for our automated future. This is the kind of futurist dream where humans will be served by robot butlers and automation will mean that human time is freed up for recreational and creative pursuits. In this future, businesses that run these machines will be heavily taxed or the machines will just be owned by the government on behalf of the people and a universal basic income will be distributed to everybody to make up for the salaries that they will no longer be able to make.
People can still choose to run a business or seek employment in the few industries where people are not obsolete to make extra income and these kinds of people will probably make up the upper middle class. Beyond this, they may even be allowed to live on less than the universal basic income allowance and invest the difference for a hope of returns in the future. But, below that, almost everybody will still live lives with higher standards of living than our middle class today thanks to the abundance caused by this new automated workforce. This all sounds great. But, let's say that a family wanted to get themselves a bigger house or just generally increase their well-being.
Working jobs that are still left for humans beings may be an option, but they are likely to be very hard to get into and possibly pretty unrewarding financially. Investing also works. By saving a little bit of their basic income and investing it into the market, they might hope to get returns and build up investment income that means they will have more income later on. This is not too dissimilar to how most people build up wealth in today's society. But, in a world with a universal basic income, there is one other possibility.
There would be nothing really to stop family groups just having lots and lots of children who would also be eligible for this universal basic income, boosting the overall income of the household. Sure, you could introduce laws that circumvent this by taking away the rights of children to get it until they turn 18, but this kind of takes away from the universalness of this universal basic income deal. And as we have seen time and time again, people are really good at finding loopholes when it comes to getting money at a government bureaucracy.
Even if we ignore this particular incentive to breed, breeding will happen because if an entire society does not need to commit to employment, then babies aren't going to be a strain on their career. Nobody will have to worry about the cost of child care. People won't need to be concerned if they are spending enough time with their children, and so birth rates will rise. It is very common in developed countries for birth rates to shrink for all of the reasons above, but all of those reasons go away when jobs are no longer a thing. So, we are likely to see an explosion of new babies. So, what? Well, this is
what. Humans are a drain on resources. This is fine for most humans today because able-bodied and or able-minded adults tend to contribute more to society than they cost, meaning that they can afford to live off their wages and still contribute to businesses or their community at large, making the world a richer place. But, in this post-automation world, we don't really need them to balance sheets, reel to real estate, or generally manage general management positions. That will all will done by machines. What won't change is that humans still need to be fed with food and housed in homes and transported with transport, and all of this puts a drain on finite resources.
The world can only support so many people with these resources it has available. And yes, technology like automation will increase the amount of people slightly, but there is still a hard cap on how much fresh water and food we can extract from our little blue rock. Space colonization could help, but it does open up the opportunity for pretty much limitless growth, and that is well into the future. We're going to have to come to grips short-term though with humans really having a negative value to society. This gets us onto the bad outcome, the outcome that is coincidentally most likely. A universal basic income structure is still in place, but it is very, very basic, just barely covering the essentials. The world is split into
two very distinct classes. Those who own the companies that own the robots that build everything and everybody else. This peasant class will be forced to take on gig style jobs to stay ahead in fields that they cannot fill with automation just yet. Having children will be highly discouraged, and things like universal basic income might only be made available to people when they do turn 18, meaning that having children will be hugely financially crippling or just left to the domain of the rich. The rich in this scenario are actually not much better off. Sure, they will sit on top of vast fortunes and have robot servants and the like, but they still have that in the first scenario as well.
This will be a world filled with violence and contemptment for these wealthy people. Picture Johannesburg, a country with very clear divide between rich and poor. Fantastic opportunities here exist for wealthy business people to live extraordinarily luxurious lives in incredible safety. But, the country has still seen a mass exodus of these wealthy individuals because people don't like living in fortresses. In this scenario, Johannesburg will become everywhere. Sounds pretty bad, right? Well, it can still get a whole lot worse.
The ugly post-automation scenario is likely to become a reality in at least a few countries where the mentality is that people should just work for a living. In this scenario, universal basic incomes won't really be there at all. Social welfare may exist, but it will likely be starved due to pure economic forces. A human that is unemployable has no economic value. It is a horrible thing to say, and most people have a hard time accepting it, but it is ultimately true. We live in a world today where most of us trade our time for money, and then we trade that money for goods and services. Pure transactional economics does not give money to people if their time is worthless.
Now, I know what a few of you are going to say. An economic system like this will never work. Businesses need customers to sell to stay in business. Who is going to keep these wealthy businesses afloat? And oh, you poor sweet summer child, this is probably just something your sociology teacher told you while gently patting your head and repeating it's all going to be all right. The truth is no. A big population is not needed to keep businesses in business. Let's take it down to the most fundamental level. People today work for business. They go to work 40 hours a week, and then they get paid money for the work that they can spend on goods to
keep them happy, healthy, and alive. They buy their food and clothes and health care from what are really just other businesses. So, at the most fundamental level, if we take away money as the middleman, businesses are really just giving people food, housing, clothing, etc. in exchange for their productive time. If their time no longer has any value, businesses will no longer provide them with housing, food, and clothing because, well, why should they? So, who will these businesses trade with then? Well, each other. There will still be a market servicing the needs of these businesses and their incredibly wealthy owners. I'm sorry to break it to you, but the average consumer is really not as important to the economic system as
they are led to believe. In this scenario, we are likely to see massive population declines. The less sinister cause of this is that people will just have less children because they won't be able to support them. And the second, much darker cause, is that people may very well just starve to death. It sounds very foreign to people who live in developed countries, but these kinds of terrible outcomes are possible as a result of full-scale automation. We have been blessed to live in a time where technology has made our work hugely productive, and this huge productivity has meant that these workers get to benefit off the wealth they have helped to create. But, it has to be asked, what does a future look
like where humans are no longer the creators of value? That was our thinking in July 2019, The Good, The Bad, and The Ugly. What we couldn't fully anticipate was how quickly the technology would actually arrive or which workers it would come for first. When we made that video, the threat felt most acute for the factory floor. For the kinds of manual and repetitive work that machines had always been most eager to take on. What happened instead was almost the opposite. When AI finally arrived, it came for call centers in Manila, data entry workers in Dhaka, and the entire outsourced service economies that developing countries had spent three decades carefully building. As it turned out, the factory floor was going to be
fine. It was the people working in an office that had something to worry about. And that changed the question entirely from what if machines replace us to what do we do now that they already are? Six years on, the thought experiment became an economic reality. Everybody has some level of anxiety over what our AI life future will look like. Somewhere between Skynet and a post-scarcity utopia, the most immediate concern for most people is that this technology will end up doing their job better than they can. So far, one side of the argument points out that big new technologies in the past have only ever made economies wealthier, and whatever jobs they replace, they end up making more better jobs somewhere else. The
other side argues that, yeah, sure. When we replaced our muscles with machinery in the past, it let us leverage our minds, which are clearly what humans have invested most of our evolutionary traits into. But, if machines replace that, what else do we have left to offer? Now, nobody can predict the future, least of all economists. But, we don't really need to, because there are certain economies that are going to see the widespread impacts of these changes well before most others. In fact, they kind of already are. In places like the Philippines and Bangladesh, the threat of AI is much more imminent. The threat to jobs, to entire industries, and to the economic growth they've spent
decades building. These economies have spent the last 30 years constructing entire industries around outsourced service work. Things like call centers, data entry, transcription, and basic software support. These jobs were once considered safe from automation because they required language skills, context, and that special human touch that machines just couldn't replace. Well, it turns out machines got a lot better at replicating that human touch. Tools like LLMs can now handle those tasks in seconds at a fraction of the cost, and these jobs, which make up a big share of GDP in many developing countries, are looking like they might be the first dominoes to fall. In the Philippines, the IMF estimates a
staggering 89% of outsourced service jobs are at high risk of being automated by AI. That's over a million people whose jobs could disappear in just a few years. In other words, AI is already making the world's richest countries even richer, and is making it harder for everybody else to catch up. And that's just the beginning of the story. Even in rich countries, AI is starting to divide the economy into those who can leverage it and those who are going to get replaced by it. The US Bureau of Labor Statistics predicts that roles like cashiers, bank tellers, postal staff, and customer service representatives are all on track to shrink. One estimate suggests 7.1 million jobs could
disappear in the next 5 years with up to 47% of current roles at risk of being replaced by AI. Of course, it's also worth remembering that companies and their investors have now plowed trillions of dollars into developing this technology, so they want to eventually see a return. As heartless as it is, cutting millions of workers off payroll is probably the most immediate way to start seeing those returns. So, there is an incentive to play up the scare campaign because what sounds horrifying to most people sounds like opportunity to those actually writing the checks. But even still, the trend lines are clear. AI is already reshaping who gets ahead, who falls behind, and most importantly, how fast the gap is
widening. So, as always, we've got some important questions to answer. Why is AI supercharging growth in rich countries while simultaneously threatening the economic survival of others? In a world where one person armed with AI can replace five people, what exactly happens to the other four? And perhaps most importantly, can workers or even entire economies adapt fast enough to survive the shift? According to the Center for Economic Policy Research, the US could see a 5.4% boost to GDP over the next decade thanks to AI-driven productivity gains. The UK, Germany, and South Korea aren't far behind with similar projections.
Meanwhile, lower-income countries are looking at much more modest gains, closer to 2.7 to 3.5%, which would be a departure from the expectations about developing countries well developing faster. The Philippines is a good example. For years, it's been one of the world's premier destinations for business process outsourcing, a $37 billion industry that includes customer service, billing, transcription, and tech support. The tech sector employs more than 1.3 million people and contributes over 7% of the country's total GDP. But here's the economic nightmare scenario. Most of these jobs are exactly the kind of repetitive, text-based tasks that large language models like ChatGPT are rapidly learning
to automate. Jobs in the Philippines are at high risk of being replaced by AI, and it's already happening. Roughly 2/3 of outsourcing companies in the country are now using AI tools to cut costs and speed up workflows. Major US companies like AT&T, Google, and Accenture outsource work to the Philippines, but if AI can perform the same task faster and cheaper, and without requiring health insurance, vacation days, or human resource departments, those jobs will be amongst the first casualties. Bangladesh is in a similar boat. Its outsourcing sector has grown to 400 firms employing over 80,000 people, but the vast majority of that work still centers around customer service, transcription, and data entry, which again is precisely the kind of job that
AI is becoming increasingly capable of automating. If AI can deliver the same quality of work at a better speed for significantly less money, there's simply no compelling economic reason to continue outsourcing. A single slot in a server rack could soon replace an entire call center in Manila or Dhaka, and that means companies could start to reshore, bringing jobs back to wealthy nations where local automation can rival offshore labor on price. That completely flips the script on the entire outsourcing model that emerging economies have built their entire growth strategies around for the past three decades, and it's a big reason why the gap between wealthy and poorer nations is set to widen after a few decades of
these economies actually slowly catching up. AI also rewards exactly the kind of specialized skills that are hardest to scale globally. Building and training large AI models requires advanced education, reliable municipal infrastructure, and access to advanced technologies. Those resources are overwhelmingly concentrated in wealthy nations. That means the most valuable AI jobs are also the least accessible to workers in emerging markets. And when workers in those countries do manage to gain access to those highly sought-after skills, they often don't stick around. Talented engineers have been recruited by global tech companies or relocating entirely to hubs like San Francisco, London, Berlin, or even centers within
China. The result is an accelerating brain drain that leaves poorer nations with fewer startups, fewer teachers and researchers, and dramatically fewer chances to catch up in the global AI race. And it's clear which countries are leading that race. In short, the countries least equipped to absorb disruption are the ones getting hit first and hardest, while the countries best positioned to benefit from AI are already pulling ahead because they control the capital, infrastructure, talent, and resources shaping the future of AI. But, AI isn't just dividing countries along economic lines, it's also creating stark divisions between the people within those same countries.
This technology is not impacting all people in the same way. It's making some workers nearly obsolete while making others far more valuable. That's because AI represents a very specific kind of capital, and understanding this distinction is crucial for predicting its economic impact. In the past, most new technologies functioned as what economists call complementary capital, meaning these were machines and technologies that made human workers more productive. For example, a combine harvester didn't eliminate farm workers, instead, it made each individual worker dramatically more efficient. Before mechanization, harvesting a single field might require 20 people working for several days, while with a harvester,
one person could do the same job in a fraction of the time. Labor and capital worked together, and as productivity increased, so did wages and living standards. Workers remained essential to the process, they just became much more productive. For many high-skilled roles, AI will become more complementary capital, boosting productivity without replacing the human worker. A financial analyst using AI to scan reports and spot anomalies can get insights faster and can focus more time on strategic thinking. A doctor leveraging AI for diagnostics can spend more time on direct patient care. In these cases, AI multiplies what skilled professionals can do and makes their expertise more
valuable in the marketplace. But, for more routine process-driven work, AI increasingly acts as what economists call substitutive capital, replacing human labor altogether instead of enhancing it. An AI-powered chatbot doesn't make a customer support agent faster, it replaces them. A sophisticated code generator doesn't assist a junior developer, it replaces them. In other words, the more capable our capital becomes, the less it actually needs human labor to function. And in the AI economy, capital ownership is more concentrated than it ever has been in modern history. Most of the major breakthroughs in artificial intelligence are coming from a handful of elite firms in the US and China.
Since 2017, the US has produced 135 large-scale AI systems. China's not far behind with 110, but the gap widens quickly. From there, the UK has managed 25 and France 24. And the companies leading the charge with these breakthroughs are experiencing exponential growth thanks to what is known as the data network effect. The more data they collect, the better their AI model performs. The better their model, the more users they attract. And the more users they attract, the more data they generate. This creates a powerful feedback loop where market power and profits concentrate in just a few dominant companies. PwC estimated that AI could add $15.7 trillion to global GDP by 2030, but 70% of that wealth is projected to go to just two
countries, the USA and China, because they own AI. In 2024 alone, over 1,100 US-based AI companies raised major funding rounds. That's more than double all of Europe combined. IBM and Microsoft alone hold thousands of AI-related patents, giving them long-term control over everything from enterprise tools to foundational models. Smaller firms, even those in wealthy countries, are becoming increasingly dependent on licensing tools and models that they didn't build and don't control. And that extends beyond software. The physical machines that power AI, CPUs and GPUs, are overwhelmingly designed and manufactured in just five countries. More than 90% of that hardware comes from the US, Taiwan, China, South Korea, and Japan. And that
means a tiny handful of countries don't just run AI systems, but also manufacture the foundational components that make AI possible in the first place. That's the reality of AI as capital. It primarily benefits those who already own the assets, while replacing those who don't. The more you can leverage AI as a productivity multiplier, the more economically valuable you become in the marketplace. But for workers in routine roles, especially those without access to retraining programs, the future is looking a far less promising. Now, even if you weren't aware of these exact figures, they probably aren't surprising. And that's exactly the point. This is a reality that people are noticing. Nearly 1/3 of Americans in a recent survey said they're fairly or
very worried about losing their jobs to automation. This isn't some hypothetical scenario we're speculating about. We've witnessed similar disruptions before. When industrial automation and large-scale outsourcing ramped up in the 1980s and 1990s, it hit manufacturing hard, especially in places like the US and Western Europe. In America alone, more than 7 million factory jobs disappeared between 1980 and 2010, and most of them didn't come back. These factory jobs may have been replacing US workers with Chinese workers, but there is no critical reason why human workers couldn't be replaced with clunkers. The Midwest bore the brunt of this economic transformation. Cities like Detroit,
Cleveland, and Youngstown were once packed with well-paying jobs in steel, cars, and textiles, but then came robotic welders, computer-run assembly lines, and cheaper labor overseas. Suddenly, those stable middle-class jobs evaporated, factories closed, unemployment spiked, and entire local economies started to fall apart. The consequences extended far beyond simple job loss. A lot of these towns saw life expectancy drop, opioid addiction rise, and schools struggled to keep up. The jobs that eventually did return often paid less and didn't offer the stability or benefits that had previously supported entire communities. The UK
experienced something similar. Coal mining, shipbuilding, and steel plants across Northern England and Scotland shut down as automation and privatization took hold. Even today, places like Sheffield and Sunderland still lag behind the rest of the country when it comes to income and social mobility. The lesson is clear. Even when the long-term picture improves, the short-term impact of technological disruption can be devastating. And once inequality takes root in an economy, it becomes extremely difficult to reverse. So, what can we actually do about this looming challenge? Because at this point, it's clear that AI is already transforming the global economy, but whether it deepens existing inequality
or helps us solve it depends on the actions that countries and individuals take in the coming years. First, the good news is we can already see what's coming our way. In lower-income countries like the Philippines and Bangladesh, the frontline effects of AI are unfolding in real time. These economies show us which jobs go first, where the risks are highest, and what happens when governments act or don't. For example, the government of the Philippines is launching a national AI strategy with the goal of retraining over a million workers by 2028. Bangladesh, meanwhile, has released a draft policy framework focused on developing AI talent, modernizing its education system, and supporting tech startups. The goal is to position Bangladesh as a competitive player in
the AI-enabled services market while safeguarding jobs through upskilling and digital inclusion programs. Whether those efforts will prove sufficient remains to be seen, but they offer a clear warning and a playbook for wealthier nations to follow. Two distinct sides of our economies need to do two things simultaneously: invest heavily into AI infrastructure and invest just as heavily into their people. This includes educational investments into computer science BS, but also the kind of skills AI struggles to automate: critical thinking, complex problem-solving, effective communication, and creative decision-making. A recent analysis of 12 million job postings in the US found that AI adoption tends to increase demand for these distinctly human skills
far more often than it eliminates jobs entirely. Building an accessible digital economy is equally important because right now nearly 2.6 billion people worldwide still don't have access to the internet. Without that basic connectivity, there's simply no opportunity to compete or even participate in the emerging AI economy. The World Bank estimates that every 10% increase in broadband access can boost GDP growth in developing countries by up to 1.4% and that's before factoring in the additional benefits that AI capabilities could provide. So, along with retraining, countries need policies that expand broadband access, reduce the cost of devices, and give more people the skills they need to benefit from AI.
Social safety nets matter, too. They function as economic buffers that give displaced workers the time and resources they need to adapt, retrain, and reenter the labor market from a position of strength. But, if AI allows businesses to grow while workers lose their income, the economy starts to hollow out. Productivity rises, but consumption falls. Innovation continues, but inequality grows, and it becomes a serious drag on overall economic growth. If we want AI to boost productivity broadly, not just corporate profits, we'll need to rethink how we design and share value it creates. That includes fundamental questions about who gets to build AI, who governs its development and deployment, and who ultimately
benefits from the massive productivity gains it generates. That video came out in September 2025. Seven months later, a team at MIT published something that reframed the entire conversation. The question of who gets hit by AI turned out to be even more complicated than we thought because we had been asking about jobs when we should have been asking about tasks. If you have read a single headline about AI in the last 2 years, you already know which jobs are supposedly under threat. Programmers are being laid off, junior developers can't find work, and tech companies are freezing hiring because one AI tool can now do the work of a dozen entry-level engineers. The discourse is loud, anxious, and honestly, a fair bit of it is justified.
But, according to a new study out of MIT, it's also almost entirely focused on the wrong thing. Most people, and most headlines, treat AI as something that replaces jobs. But, that's not how it works. More often, AI replaces the tasks inside them. A lawyer doesn't disappear overnight, but the hours they spend reviewing routine contracts might quietly shrink. A journalist keeps writing, but the time spent researching background, pulling quotes, and fact-checking starts to compress. And while that might sound like a softer version of the same story, it definitely isn't because our entire economic system, the way we measure work, track productivity, and plan for the future is built around jobs, not tasks.
GDP, unemployment figures, and wage data were designed to count jobs and people, and they do that reasonably well. But, they were never built to look inside a job and ask which parts of it AI can already technically perform. So, the change doesn't show up where we're looking for it. By the time the disruption shows up in the official numbers, it's already well underway, and every plan governments have made to prepare their workforces has been built on instruments that are pointed in the wrong direction. It's a bit like trying to navigate a new city using a 30-year-old map. The streets look familiar, but nothing is quite where you'd expect. This is exactly why a team at MIT decided to build something new. Not another prediction of which
jobs will disappear, but an actual map of where AI capabilities and human skills currently overlap, weighted by the economic value of that work. They called it the iceberg index, and the name turns out to be exactly right. When you measure the work AI can technically perform across the tech sector, it accounts for about 2.2% of total US labor market wage value, roughly $211 billion. That's the visible tip of the iceberg, but when you apply the same methodology to the whole economy, the number jumps to 11.7%, roughly $1.2 trillion, and five times larger. That's the part underwater, and it includes highly educated, well-paid professionals in industries and sectors that haven't generated a single anxious headline about AI.
So, as always, we've got some important questions to answer. What exactly is the iceberg index, and why does asking about skills instead of jobs change everything? What does the index actually reveal about what's going on below the surface, and how much of the story we've been missing? And finally, what happens to the workers and industries that AI simply cannot touch? And why is that answer a lot stranger and a lot more expensive than most people expect? Every major economic transition in history eventually forced economists to build new tools to measure what was actually happening, because the old ones simply weren't designed for the new reality. In the industrial era, output
per hour had to be invented because nobody had a way to capture physical productivity at scale. In the internet era, GDP ran into a rather embarrassing problem, which is that it could measure the value of encyclopedias sold in shops, but couldn't measure the value of Wikipedia, which replaced them altogether. So, the US Bureau of Economic Analysis eventually had to build an entirely separate accounting framework, the digital economy satellite account, just to capture the value of services that existed outside normal market transactions. MIT is arguing that the AI era calls for a similar overhaul. Intelligence is now a shared input between humans and machines, and the tools we inherited were simply never
designed to measure that. The Iceberg Index is MIT's attempt to build something that is. So, how did they actually do it? Well, they started by creating a digital representation of 151 million American workers spread across 923 different occupations and 3,000 counties. To map the skills each occupation requires, they use something called O*NET, a database maintained by the US Department of Labor that breaks down hundreds of occupations into their actual component skills and tasks. Things like analyzing data, critical thinking, interacting with computers, programming, and coordinating with others. And in the database, each of these skills comes with an importance rating and a difficulty level built from
surveys of real workers doing those jobs. So, instead of saying there are 4 million accountants in America, the database is saying, "Here are the 37 specific skills accounting work requires, here is how important each one is, and here is how difficult each one is to perform." Which is either a very thorough way to think about employment or a very bureaucratic one. But, either way, it turns out to be extremely useful. Then, they did the same thing for AI, cataloging more than 13,000 real production-ready AI tools that exist and are being deployed inside companies right now. From coding assistants and document processes to financial analysis software and workflow tools on platforms like Zapier. And crucially, they put every single one of
those tools through the same O*NET skill taxonomy. So, you end up with a profile of what each AI tool can do, expressed in exactly the same language as the human workers. Now, you have two maps using the same framework. And for the first time, you can make a genuinely apples-to-apples comparison between what human workers actually do and what AI systems are technically capable of doing right now. The result is a single number for each occupation, a percentage that measures how much of the wage value inside that job AI can technically perform. And the word wage value is actually the most important design choice in the
whole study, because rather than asking how many accountants might lose their jobs, they asked how much of the economic value that accountants produce comes from skills AI can already perform. An accountant might spend 60% of their time on document processing and data entry, and 40% on complex judgment calls and client relationships. Those are very different things, and treating them as equivalent would tell you almost nothing useful, because automating 60% of someone's time doesn't mean automating 60% of their value. Weighting by wage value, on the other hand, means the index reflects where the actual economic exposure sits. Now, there are a few things the iceberg index deliberately does not do, and it's worth
being clear about them. It doesn't account for physical robotics. For now, this is purely about digital and cognitive AI tools. And it measures technical capability only, not outcomes, which means it doesn't predict job losses or forecast when any of this will actually happen. What companies do with this information, when they do it, and whether regulators allow it, none of that is in the model. Think of it less like a weather forecast and more like an earthquake risk map. It tells you exactly which buildings are sitting on a fault line, not if or when the tremor will hit. So, what does the map actually show? Before we get into the findings, let me tell you about the Economics Explained newsletter. This channel covers big
stories worth 15 minutes of your time, but there's a lot of important economic stuff happening every week that doesn't quite make the cut for a full video. So, we write about it instead, covering things like the businesses selling fake academic research that's corrupting trillions in government spending, why three companies are quietly profiting from pharmacies closing, and how a water shortage in Mexico sent cilantro prices up 400%. It goes out once a week, it's completely free, and it won't waste your time. The link is in the description below, or scan the QR code on screen. Now, back to what the iceberg index actually found. It starts with the part everyone already knows about. In 2025
alone, more than 100,000 job losses were linked to AI restructuring. AI is already writing more code every day than every human developer on the planet combined, and we already know that when you measure the wage value of the work inside those tech jobs that AI can technically perform, it comes to about 2.2% of the entire US labor market, roughly $211 billion across 1.9 million workers. This is where almost every headline, every policy paper, and every anxious conversation about AI and jobs has been focused. But, 2.2% is a misleading place to stop. 2.2% is just the visible tip of the iceberg, but the capabilities driving that number, skills like document processing, routine
analysis, and data handling don't belong just to tech. They show up across hundreds of occupations that would never appear in a headline about AI layoffs. The same capabilities that make a coding assistant useful to a software engineer overlap heavily with the work of a financial analyst, an HR coordinator, an insurance claims processor, and a legal secretary. So, when you run the index across all the skills, that's where you see the number jump from 2.2% to 11.7% of the total US labor market, or roughly $1.2 trillion in wage value.
This means the anxiety about AI that's been dominating headlines for 2 years has been aimed at roughly 1/5 of the actual problem. The other 4/5 have been sitting on a fault line that no government, no company, and no individual worker has been preparing for. And the people sitting on it aren't who you'd expect. According to a separate Anthropic study tracking actual AI usage in professional settings, the most exposed group earns 47% more on average than the least exposed, is nearly four times as likely to hold a graduate degree, and is 16 percentage points more likely to be female. In plain terms, anyone whose working day is built primarily around reading, writing, analyzing, and summarizing information.
People who, by any reasonable measure, did everything society told them to do, and did it well. Now, the gap between what AI can technically do and what it's actually doing in practice is still enormous. And for the moment, that gap is what's keeping much of this exposure theoretical rather than real. For computer and math workers, AI is theoretically capable of handling around 94% of their tasks, but in observed professional use, it's currently doing about 33%. And a similar pattern shows up in legal work, in architecture and engineering, and across many other professions. The technical capability is already there, but right now it's being held back by regulation, integration challenges, and
the simple fact that most organizations still require a human to check AI's work. These are all friction points that may resolve themselves as technology matures. And while the full exposure hasn't landed yet, the leading edge of it is already visible in the hiring data. IBM replaced a chunk of its HR department with AI tools, while Salesforce stopped hiring engineers and lawyers because, as its CEO put it, AI can do the work. Entry-level employment in AI-exposed occupations has already dropped 14% compared to the pre-ChatGPT era. And that's probably just the beginning because job postings fall before employment. And entry-level job postings across the US have dropped
35% since January 2023. Now, if you had to guess which states are sitting on the biggest fault lines, you'd probably say California, Washington, and New York. But, you'd be wrong. According to the Iceberg Index, South Dakota, North Carolina, and Utah show higher exposure values than California or Virginia. And yes, I know those aren't exactly the states generating anxious op-eds in the New York Times. The problem is, their economies happen to be heavily concentrated in administrative and financial work, which are the exact skills sitting in that gap between what AI can technically do and what it's currently doing in practice.
California's workforce is diversified enough that the exposure spreads thin, but in states built around finance and back office services, the vulnerability concentrates. Tennessee makes this point most starkly. Its tech sector exposure is 1.3%. Nothing that would trigger alarm bells in any standard workforce planning model, but its Iceberg Index is 11.6%, which means the white-collar workforce keeping Tennessee's factories running is 10 times more exposed than the tech sector everyone's been watching. Ohio and Michigan follow the same pattern. These states have spent years preparing for robots to take over the factory floors, but the white-collar disruption is arriving first, which brings us to
the most important finding. If the exposure is this widespread and this geographically distributed, why aren't the people responsible for preparing the workforce seeing it? A big part of it is simple. The tools they're using cannot see this kind of risk. GDP, per capita income, and unemployment, the standard metrics, explain less than 5% of the variation in Iceberg Index scores across states. In some cases, the relationship even flips, meaning states that look safest by conventional measures aren't necessarily the least exposed. At the same time, the states that look most vulnerable aren't necessarily the most at risk. That means
the billions being spent right now on workforce preparation may be systematically aimed at the wrong places entirely. But that's only the picture for workers inside that 11.7%, the ones exposed to AI. What about the workers who aren't? Because according to Anthropic's research, about 30% of the workforce has essentially zero AI exposure. This includes cooks, mechanics, nurses, plumbers, bartenders, childcare workers, people doing physical, relational, hands-on work that no language model can replicate. Surely those workers are fine, no? Well, sort of. You see, there's an economic pattern that's been quietly working against these workers for decades, long before anyone had heard of a large language model. To understand the pattern, you need to go back to 1965
when a Princeton economist named William Baumol and his colleague William Bowen noticed something odd about the performing arts. A string quartet performing Beethoven in the 19th century required four musicians and about 25 minutes. A string quartet performing the same piece one century later required the exact same number of musicians and lasted about the same time. Nothing about the performance had gotten more efficient. No technology had come along to speed it up, and yet the cost of putting on that concert had risen dramatically, dragged upward by wages rising everywhere else as manufacturing, agriculture, and industry got more and more productive. As factories and other industries became dramatically cheaper and faster, they
could afford to pay their workers more, which pushed wages up across the whole economy, and musicians had to be paid competitively to attract talented people into the profession, even though the output, four people playing for 25 minutes, never changed. So, the cost per performance kept climbing. Economists call this Baumol's cost disease, and the clearest way to see it is to look at what happened to prices over the last 50 years. The things that got more productive, like electronics and computers, got cheaper, while the things that couldn't get more productive, like child care, education, haircuts, health care, and legal services, kept getting more expensive. These are all industries
where the work is stubbornly human. You can't make a nurse see patients faster or a plumber fix a pipe remotely. So, the only way to cover the rising wage bill is to raise prices, and AI is about to make this much worse. If the iceberg index is right and cognitive and administrative work is about to get dramatically more productive, then the Baumol effect will accelerate. Over the coming years, a financial analyst processing documents with AI assistance might shrink a day's work into an hour, and a software engineer with the right tools might do the work of three people. But, the nurse will still need the same amount of time per patient, and the plumber will still need to physically be there. Their output
doesn't scale with AI, so the relative cost of their work keeps rising, pulled upward by a productivity surge happening everywhere around them. This matters because most of the work AI can't touch is essential. People don't pull their kids out of school because costs went up or skip the plumber when the pipes burst. Health care, education, elder care, and skilled trades aren't things people can just stop using when prices rise. And most of them are either funded or subsidized by governments that are already stretched thin. So, workers who are safe from AI disruption may find themselves in industries that governments will increasingly struggle to fund. Now, I want to be clear about something. The Baumol argument is a
reasonable extrapolation from what the study found, not a finding in the study itself. Because the iceberg index tells you where the skill overlap sits right now, not what firms will do about it, how fast governments will respond, or which workers will successfully retrain. Those outcomes depend on decisions that haven't been made yet. And look, transitions like this have historically created as many new roles as they have disrupted. The transitions that went best were the ones where people could see what was coming clearly enough to prepare. Today, governments and companies are making billion-dollar decisions about AI using tools that can't see 95% of the problem they're trying to measure. The map most people
are using was drawn for a different economy. The Iceberg Index is an attempt to draw a better one. Whether anyone actually uses it is, as always, another question entirely. But if they don't, the cost is a two-speed economy. One half dramatically more productive, the other steadily less affordable, with the fiscal pressure landing hardest on the exact services societies can't function without. Nobody can predict exactly how that plays out, least of all economists. In the month since our MIT video came out, employment among software developers under 25 has already fallen 20% from its 2022 peak, as the entry-level tasks that used to be the way in are being handled by something else. Most organizations are already
using generative AI in some capacity, seven in 10 at last count. And three-quarters of the economic gains from all of this are being captured by just 20% of companies. What I got right in 2019 was the core tension that automation creates winners and losers, and that the economic system has never been particularly good at distributing those outcomes evenly. What I didn't anticipate was how quickly the theoretical would become the actual and how far outside the factory floor the disruption would reach. Nobody can predict the future, least of all economists. The best we can do is keep updating the model. Thanks for watching, mate. Bye.
