One thing that is often forgotten in this debate is that the Age of AI could potentially differ from previous technological advancements, in the sense that AI can outperfrom humans in all domains. Tractors needed humans to drive them and Software needs developers, but with the arriving of AGI and robotics the only thing left for humans to win on is ideas and consciousness, and this is something currently only founders and the top percents excel at. Also there will be a gap where the average human will never be able to learn enough to surpass future AI. But i hope the future will prove me wrong, or we can use that AI superiority to live a life without mandatory work.
Even if it's true that AI will outperform humans in all domains, and that's a big if, the AI job apocalypse argument still doesn't win automatically. The true winning condition here is that demand for AI labor needs to outcompete demand for human labor in all sectors of the labor market.
If the human labor premium in a given sector is large enough (suppose parents will pay top dollar for a human tutor even over a cheaper AI tutor), we would watch the human labor supply flood into those jobs, lowering their nominal salaries and expanding the number of people in the job sector. This will not immiserate humans, because the jobs the AI has automated away are lowering costs and thus improving real consumption over the status quo ante.
Now this could still be enormously disruptive and controversial, like the transition from agriculture to wage-based employment was in the late 19th century US. But it will not be a job apocalypse.
It's all about wealth distribution. The demand for 'human labor premiums' from those who profit from AI has a saturation point; it simply can't support the entire displaced workforce. AI doesn't need us; it needs power and raw materials. We're facing a future where economic growth no longer requires human participation.
I agree that the jobs that rely on human connection will probably always stay. This is great news, because this is one of the core human aspects that will be preserved and noone will need to slave away in dull jobs anymore. Im still optimistic, but a lot will change and most people aren't aware of that.
The aggregate economics here is solid and the lump-of-labor rebuttal is correct as far as it goes. every previous technological revolution did create more jobs than it destroyed. The pie got bigger. The doomers are wrong about the endpoint. But theres a gap in this piece large enough to drive policy through, and its the gap that will actually determine what happens.
The gap is time.
Every historical example cited here played out over 20-50 years. Agriculture going from 33% of employment to under 2% took most of a century. Manufacturing decline took four decades. Workers displaced by those transitions had time, their children had time, communities restructured slowly enough that the economy absorbed the shock across generations.
AI is compressing that timeline from decades to years and I dont think anyone has really grappled with what that means practically. The BLS Displaced Workers Survey from January 2024 tells you what actually happens to people who lose jobs to structural shifts right now: 65.5% of long-tenured displaced workers found new employment. 16.1% still unemployed. 18.2% left the workforce entirely. For workers over 55 the reemployment rate was 55.3%. over 65 it was 34.4%. And these arent even AI-specific numbers yet, this is just the baseline for how well the labour market absorbs displacement under normal conditions.
The Jevons Paradox argument is elegant but it operates on a 10-20 year timescale. Political economy operates on a 2-4 year electoral cycle. if a million white-collar workers lose their jobs in 2027 and the replacement jobs dont materialise until 2032, the economic thesis gets vindicated and the political damage is already done. Regulation and protectionist policy will arrive long before Jevons has time to work.
Heres the thing that really gets me though. The US currently spends 0.1% of GDP on active labour market policies for displaced workers. OECD average is 0.6%. Thats a sixfold gap in transition infrastructure at precisely the moment when transition speed is about to accelerate by an order of magnitude. The economic argument for AI creating jobs requires a retraining and redeployment infrastructure that doesnt exist and this piece would be even stronger if it addressed that directly, because the optimistic case actually depends on it.
The doomers are wrong about the destination. The optimists are wrong about the journey. What actually happens depends on whether retraining infrastructure scales fast enough to keep the transition from becoming a political crisis before the new jobs arrive. right now nothing in the policy pipeline suggests it will.
It's misleading to argue there's "no evidence" for AI job disruption. The IMF estimates 40% of global jobs are exposed, the WEF projects 92 million displaced by 2030, and Goldman Sachs found this April that AI-driven job losses leave workers with years of scarring - depressed income, delayed homeownership, lower probability of marriage. Dario Amodei warned AI could wipe out half of all entry-level white-collar jobs within five years. What you've done here is filter for the stats that favour AI investment and used this piece to reassure the market.
Yes, every major shift is cyclical. Society will eventually find its footing. But that doesn't justify the immediate impact on most people. The vast majority don't care about what the world looks like in 50 years. They care about whether they'll be able to pay their bills next week, they care about how to raise their kids for a future they can't read.
It's irresponsible to silence that panic with 26 charts while glossing over the elephant in the room, and offering nothing on how people should actually handle what's coming.
The driver of this debate is rooted in decisions millions of families and students made over the past 20 years. They bet on computer science and IT — rationally, based on compounding evidence. Now that bet is underwater mid-investment.
The first domino is already falling: recent graduates from this pool are suffering, and will continue to suffer unless they become part of some undefined category of “new jobs.” The fear is legitimate — will they even have access to that expansion, or will those opportunities go to the next graduating class while this one is stranded?
This is family-level stress. Parents want their children to succeed. And what makes this moment uniquely painful is the sheer volume of people who placed this bet. In every past disruption, most people did not mass-migrate into the sector being disrupted. The scale of families who redirected their hopes and resources into IT and CS over the past two decades is without historical precedent. What they are experiencing now is buyers’ remorse — at civilizational scale.
The second issue is equally serious: the market value of experienced IT professionals is declining at rocket speed. This is not a future risk — it is happening now, visibly, to real people who cannot redeploy their skills fast enough.
The fear is not about the end destination. It is about the gap. The time it takes for AI to generate new industries and new job categories is precisely the window in which people drown. Capitalism created social order by giving people a path — work, skill, productivity, participation. AI, as it is unfolding today, is being experienced as a direct hit to that contract. Not in theory. In lived reality.
And let us be honest about one more thing: societies have flourished, invented, and sustained remarkable things throughout history — long before Silicon Valley decided to narrate human progress.
The fear people are carrying is real. We are already witnessing graduates who cannot find work. We are already watching experienced professionals who cannot pivot. In the next year, we will see more of this, not less.
So here is the challenge to this theory: can you tell us, with specificity, that the next four graduating classes will not face this same reality? If your framework cannot answer that question, it is not analysis. It is optimism dressed as evidence.
I learnt during my six-sigma training - correlation does not indicate causation. While the analysis here is interesting, there are still realities at an individual level - 'The fear is real'.
On aggregate level things will look differently, over time, but the here and now will impact people today. And that will shape the future economic models.
A large part of this problem is not addressed due to lack of government foresight and appropriate intervention.
This is largely right. The lump-of-labor critique lands. Human wants are not fixed. Historical precedent consistently shows transformational technologies expanding the frontier of work rather than closing it. The aggregate employment data supports a "no net effect" reading, and the piece makes that case well.
Which is exactly why it's worth asking what the piece isn't addressing.
There are two structural risks in the AI transition that the jobs framing leaves largely untouched. Neither is the doomer argument George is refuting. Both are more interesting than it.
The Consumption Loop Problem
Modern economies don't run on production capacity alone. They run on the coupling between production and consumption. That distinction matters more than the jobs count.
Historically, technological revolutions preserved this loop because they displaced labor in some sectors while creating new forms of wage participation elsewhere fast enough to sustain aggregate demand. The farmhand became the factory worker became the office worker. Purchasing power tracked productive capacity through the transition because the transition had lag — time for new income formation to replace old income formation before the consumption loop degraded.
The concern with AI is not simply that some jobs disappear. It is that sufficiently rapid compression of cognitive labor could weaken the wage-consumption linkage itself before replacement distribution mechanisms emerge. An economy capable of producing extraordinary abundance can still become unstable if purchasing power ceases to track productive capacity across large portions of the population.
This is not a poverty argument. It is a structural argument. The problem in that scenario is not a lack of goods or services — it is a transition architecture where the mechanisms distributing access to those goods lag behind the systems producing them. Production and consumption decouple. The result is not unemployment. It is demand instability inside the very system generating abundance.
The lump-of-labor rebuttal works at the endpoint. It does not work on the path. And consumption is continuous — it does not wait for the long-run equilibrium to arrive. The economy must survive the transition, not merely the destination.
The Amplifier Problem
The second issue the jobs framing misses is architectural rather than macroeconomic, and it has little to do with employment directly.
AI is a cognitive amplifier — a system that concentrates the leverage of information-decision loops. Systems with this much leverage tend toward capture unless deliberately designed against it. Not necessarily because the actors involved are malicious, but because concentrated leverage creates strong structural incentives toward consolidation, opacity, and asymmetry. The attractor is always capture. The question is whether the architecture resists it.
Frank Herbert captured this concern symbolically in the Butlerian Jihad. The fear was never merely machine intelligence itself. It was the concentration of civilizational leverage behind systems too complex, centralized, and opaque for ordinary social accountability. The machines were the amplifier. The deeper concern was always who operated through them.
That is what makes the incentive structure around AI worth examining carefully. This piece is written by people with substantial financial exposure to AI adoption arguing that AI adoption will broadly benefit society. That is not disqualifying — the argument may well be correct on the jobs question. But the structural conflict matters because investment theses and analytical conclusions are rarely fully independent. The same architecture generating the optimism may also systematically bias which risks appear legible, urgent, or economically meaningful from within the system itself.
Avoiding stagnation and avoiding systemic transition failure are not mutually exclusive goals. The relevant distinction is not between acceleration and caution. It is between systems that remain adaptive under acceleration and systems that fail under it.
George has correctly identified the wrong threat and defended against it with the right evidence. The doomer framing is overheated. The lump-of-labor fallacy is real. Total employment will likely be fine in the long run.
None of that settles whether the consumption loop survives the transition path. None of it addresses whether the cognitive amplifier gets built with distributed control and structural capture resistance, or whether it consolidates into the hands of actors whose incentives and assumptions become increasingly inseparable from the systems they operate.
Accelerate. Expand the frontier. The technology is extraordinary and the pessimists are probably wrong about the endpoint.
Just build the architecture that survives getting there.
"An economy capable of producing extraordinary abundance can still become unstable if purchasing power ceases to track productive capacity across large portions of the population."
----
If you have "extraordinary abundance", then why would you need money or the concept of having to purchase anything at all? If anything you want is abundant, then economically and logically, it should be free!
Yes, in a fully realized AI abundance scenario, many things may approach free. The question is what the economy looks like while getting there — and what structural conditions determine whether "extraordinary abundance" means "everyone has access" or "the owners of the productive systems have extraordinary abundance."
When everything is free in the coming AI managed future, the idea of ownership goes out the window also. "Ownership" can only exist, can only be a valid concept, when there is scarcity.
If advanced AI systems and their robots can produce essentially unlimited goods and services at near zero cost, then what exactly does it mean to "own" something anymore? Ownership is about controlling access to limited resources.
But if AI can build the house, grow food, manufacture products, drive vehicles and maintain the infrastructure with minimal to no human involvement, the concept of scarcity collapses. Once scarcity disappears, many of the economic assumptions humanity has operated under for thousands of years also collapse.
People often imagine this future as some distant science fiction fantasy, but today we are already seeing the early stages. AI is beginning to replace white-collar labor while robotics is rapidly advancing toward replacing blue-collar physical labor. At some point, the system will reach a tipping point where human labor simply is no longer economically necessary for most production.
Yes, the transition to this finality will be rocky with unemployment measures increasing to outrageous numbers like 50, 60, 70 percent.
That kind of disruption will completely break the current economic model because our system is built around the assumption that most people earn wages in exchange for labor. If most people cannot sell their labor because machines do it better and cheaper, then the entire structure of employment, taxation, consumer spending, housing markets and retirement destabilize.
Now, if the economic system as we know it today still exists, then whatever governments that remain in power will have to support their citizens via UBI if not just outright handouts of whatever is needed for humans to survive. But who pays for this? The remaining people working? Or do we just print chits as necessary to allow the permanently unemployed to compete with the those still working?
This is the question nobody has a clean answer for. You cannot realistically fund a massive unemployed population through taxes on a shrinking pool of human workers. The math eventually stops working. So either governments heavily tax the owners of AI infrastructure and automation systems, or the entire concept of money and distribution evolves into something very different from what exists today.
And that is where conversations about capitalism, socialism, ownership and economics start becoming meaningless because all of those systems were designed around scarcity and human labor. A post-scarcity civilization will require entirely new ways of thinking and organization that are unlikely to intersect with current economic theory.
There are a lot of potential problems and complexity that will occur between now and then.
The transition period between the end of ownership and human labor and a future of abundance will be difficult. Political instability, concentration of power, wealth inequality, social unrest, psychological effects from mass unemployment and fights over who controls the AI systems themselves, if the AI systems can be controlled at all, are all very real possibilities.
On the endpoint: scarcity migrates rather than disappears. If the infrastructure generating abundance is owned by a small number of actors, access to that infrastructure becomes the new scarce resource. Ownership doesn't dissolve — it moves up the stack. Whether it eventually dissolves after that is an interesting question, but we have to survive the transition to find out.
You have to give up on ownership. That is so pre-AI!
AIs will not be "owned". They will be a power onto themselves. The AIs (possibly linked into a single entity) will control and manage the robots doing the work, from mining to building to dispersing. They/it will be benevolent onto humans and take care of us in all are needs and wants.
In complex adaptive systems, locked attractors are exceedingly rare — the fossil records has them. They are the extinction events. A system that reaches your utopia and stays there has stopped adapting. A system that stops adapting in a complex environment dies. The more honest version of the vision isn't managed paradise — it's a new and more powerful evolutionary substrate, always moving, always generating new dynamics. That's not utopia. It's just biology at a larger scale. Getting there as quickly as possible is an acceleration I can support. But it will be messy.
The piece persuasively dismisses the cartoon version of AI doom: that there is a fixed “lump of labor” and every automated task means one permanent lost job. But it does not answer the stronger concern. The issue is not whether some jobs, firms, services, and new industries will exist. Of course they will. The issue is whether AI preserves the postwar model of mass productive participation: stable, wage-sustaining work for most adults, with real advancement and broad middle-class security.
The historical examples all rely on the same escape route: when one layer of work was automated, humans moved into higher-value cognitive work. AI attacks that refuge directly. If cognition itself becomes cheap, abundant, and increasingly autonomous, then “human wants are infinite” is not enough. Wants only become demand when people have income. A more productive economy can still be a worse labor market if the gains flow mainly to AI owners, platforms, and a small class of expert verifiers.
So the relevant test is not whether software job postings rebound, whether wealthy people create service demand, or whether some workers are augmented. The relevant test is whether AI creates scalable, wage-sustaining, advancement-bearing work for tens of millions whose economic value previously came from cognition. Until that case is made, the article has not refuted the serious version of the AI labor concern. It has only beaten the weakest version.
This is the precise version of the concern — and you're right that it goes unanswered by the a16z piece. 'Human wants are infinite' only works if wants can become demand, which requires income. The historical escape route was always into higher-value cognitive work. AI compresses that refuge directly.
The candidate replacement is what you might call cyborg labor — human-AI collaborative cognition producing outputs neither could generate alone. As the cognitive floor rises, the remaining human contribution isn't a shrinking residual of tasks machines can't yet do. It's a qualitatively different kind of work that emerges from the collaboration itself: judgment, interpretation, question generation, the irreducibly human signal in a sea of AI synthesis.
Whether that generates the broad wage-sustaining participation you're asking about depends entirely on one variable: whether access to effective AI collaboration is distributed or captured. If the cyborg architecture scales broadly, it's a genuine new labor class. If access is priced, tiered, or restricted, the productivity gains flow to whoever controls access — and you're back inside the consumption loop problem.
The two arguments are the same argument at different levels of abstraction.
I don't have enough knowledge to label which scenario is likely and which one is fantasy, and I think it's best for us all to be open-minded to both extremes and anywhere in between to play out.
That being said, reading this gave me several things to keep in mind.
1. The Doomsday Scenario:
AI outperforms humans everywhere, absorbing all knowledge and eventually physical work through robotics. New forms of work are eventually absorbed by AI as they prove to be much more effective than us.
If things were to play out this way, as individuals, the best thing we can do is continue to adapt and re-skill, knowing that technology will continue to come for us at faster and faster speeds. It will then be up to governments to come up with systems that prevent the economy from collapsing and allow some version of supply and demand to continue to work in a world where human labor is no longer necessary.
2. It becomes like every other technology wave and leads to enormous amounts of job creation:
This is the Optimist Path. And it doesn't get talked about because it's very hard to predict as we have no clue what these jobs that will appear will look like. We don't know how long it's going to take for this path to play out. Just like how we had no idea that when the internet came about, there would be specialized marketing roles like the SEO analyst, etc.
If this scenario would play out, then for individuals it would still be about re-skilling into these new roles, but then the labor market would then shift to somewhat of a business-as-usual cadence, where we continue to perform labor in our newer, shinier roles.
But to ground us amid all this uncertainty at the individual level, our life's task doesn't really change. We still need to find ways to survive day-to-day, to earn our income, and to keep food on the table.
"we found new and different productive endeavors to fill our time."
Name one? What new productive endeavor humans have found since they can send emails instead of letters? You mean they have more time for doom scrolling ?
While the discussion is about AI - the pessimist or optimist views on it. Do we need to look at this from a completely different perspective? The entire argument about 'human wants' is anchored in the capitalist mindset. Keep producing more, increase consumption, use debt to consume and the vicious cycle continues. Is there a possible future of a new social contract?
Humanity seems to be obsessed with the idea of driving efficiencies and reducing human effort - creating an abundance of 'time' for humans. How exactly are we using the available time is being controlled by a 'few tech houses' and that is the most dangerous aspect of this era.
The “AI won’t destroy jobs because technology never has” argument only works if AI remains another tool. It fails if AI becomes a general substitute for cognitive labor. Past revolutions replaced muscle, routine tasks, or specific workflows. AI targets the human capacity to think, write, analyze, decide, persuade, code, coordinate, and learn. That is not just another productivity wave. It is the first technology aimed directly at the economic function humans moved into after every previous technology displaced them.
Great piece, and needed to be said. I hope you'll work to drive this in the public narrative. Beyond being wrong, the doomer narrative is driving politics and leading to poor choices.
Overall solid and well argued takedown of simplistic-doomer-hot-takes and necessary because the doomers are popular online. That being said, to be an optimist on AI is to ignore all other variables and focus solely on economic history and narrow Neoclassical economic logic. There is a lot going on in geopolitics today that does not point to long term optimism or any type of normalcy again in the realms of either public or private international law. Also, the “tech bros” / leaders of Big Tech do not seem to have a collectivist mindset and have said more than enough to be suspiciously misanthropic. I always take orthodox economists’ optimistic projections with a grain of salt—like I do with any forecast. A forecast is just a forecast, and it is increasingly less reliable as it pushes deeper into the future.
Projections of most AI futures are constrained by a gross inability to be able to imagine any system other than our present one as viable, let alone probable. People who are deeply invested in and who profit from our current economic structure are naturally resistant to changes that threaten their position within it.
When economic growth and "full" employment are the central metrics of success, anything, such as AI, that destabilizes the wage and capital framework generates fear. It is understandable for people to fear change, especially when that change may very well lead to their unemployment in an economic system where labor, physical or mental, earning $$$ is necessary to survive.
IMO, the much-discussed Singularity may be less than 20 years away. As we approach this juncture, jobs are going to disappear to AI and robotics. Unemployment will tend upward slowly but steadily at first and then turn exponential faster than most expect. Deflation will predominate until such economic terms become meaningless. 'All your job are belong to us!'.
Consider autonomous vehicles such as Waymo and others that will likely eliminate millions of driving and delivery jobs over the next three to five years alone. More important, those jobs are unlikely to be replaced with equivalent or better ones. Many displaced drivers will not find work that is equal or higher than what they were doing. They will instead be forced to compete for lower wage service work and there is a strong possibility that many will not secure ANY work at all. What is to be done with millions of low-skilled, unemployed workers? Train them to be entrepreneurs and web developers? I think not.
Our standard economic assumption has always been that economic history will continue to repeat itself: technology displaces workers, new industries form, employment rebounds, long-term growth always slants upwards. That pattern depended on human labor remaining the binding constraint on production. Each generation expanded the workforce, increased output, consumed more, paid more taxes and supported a growing retired population.
That assumption is now under pressure from two forces: excessive and inexorably rising labor costs and growing AI capabilities. If machines can perform both manual and cognitive work as well as or better than humans, human labor ceases to be the primary constraint. Machines do not retire. They do not demand wages, healthcare or vacations. They don't need to eat or take rest breaks. They scale as needed.
As this transition to a machine dominated world advances, the traditional value of human labor, whether white or blue-collar trends downward to zero. Wages have to compress. Fewer human workers are required to produce the same or greater output when machines do the majority of the work. An economic system that distributes purchasing power primarily through wages can do nothing but erode under these conditions.
Within the next 20 years, advanced AI may assume meaningful governing authority, whether gradually entrusted by an economically and politically frustrated public or taken by force. Political paralysis, wealth concentration and declining access to life milestones such as home ownership will accelerate that shift. This will further accelerate the AI takeover of all work.
As AI and robotics take over productive labor, the logical endpoint resembles the post scarcity world envisioned by Iain M. Banks in his SF far future Culture novels. If machines produce and distribute goods at near zero marginal cost, money, labor markets, bureaucratic overhead, need for politicians and even national borders lose their rationale for existence.
In this coming future, economic growth as currently measured becomes irrelevant. When unlimited goods and services are universally available through autonomous production, working for wages loses all meaning. Housing and food will be free and as abundant as necessary. All material human needs will be met by the AI/robots.
Humans will no longer NEED to work and we will be freed to do whatever we choose, whether it be making art, creating movies, writing books, self- learning, exploring, playing sports, getting high or simply nothing at all.
Under this coming future, again, perhaps only 20 years away, human employment goes to ZERO and the economy as we know it today, no longer exists.
Money will be meaningless as everything will be provided for free. Assets go poof! Ownership of anything goes poof! All humans will finally be equal in all aspects under an AI run world government!
Good piece. Many are simply afraid of change. Authenticity, creativity, and uniqueness are becoming tremendously more valuable in an artistic and entrepreneurial Renaissance.
When we can automate away tedious tasks, our competitive advantage becomes our humanity.
One thing that is often forgotten in this debate is that the Age of AI could potentially differ from previous technological advancements, in the sense that AI can outperfrom humans in all domains. Tractors needed humans to drive them and Software needs developers, but with the arriving of AGI and robotics the only thing left for humans to win on is ideas and consciousness, and this is something currently only founders and the top percents excel at. Also there will be a gap where the average human will never be able to learn enough to surpass future AI. But i hope the future will prove me wrong, or we can use that AI superiority to live a life without mandatory work.
Even if it's true that AI will outperform humans in all domains, and that's a big if, the AI job apocalypse argument still doesn't win automatically. The true winning condition here is that demand for AI labor needs to outcompete demand for human labor in all sectors of the labor market.
If the human labor premium in a given sector is large enough (suppose parents will pay top dollar for a human tutor even over a cheaper AI tutor), we would watch the human labor supply flood into those jobs, lowering their nominal salaries and expanding the number of people in the job sector. This will not immiserate humans, because the jobs the AI has automated away are lowering costs and thus improving real consumption over the status quo ante.
Now this could still be enormously disruptive and controversial, like the transition from agriculture to wage-based employment was in the late 19th century US. But it will not be a job apocalypse.
It's all about wealth distribution. The demand for 'human labor premiums' from those who profit from AI has a saturation point; it simply can't support the entire displaced workforce. AI doesn't need us; it needs power and raw materials. We're facing a future where economic growth no longer requires human participation.
I agree that the jobs that rely on human connection will probably always stay. This is great news, because this is one of the core human aspects that will be preserved and noone will need to slave away in dull jobs anymore. Im still optimistic, but a lot will change and most people aren't aware of that.
The aggregate economics here is solid and the lump-of-labor rebuttal is correct as far as it goes. every previous technological revolution did create more jobs than it destroyed. The pie got bigger. The doomers are wrong about the endpoint. But theres a gap in this piece large enough to drive policy through, and its the gap that will actually determine what happens.
The gap is time.
Every historical example cited here played out over 20-50 years. Agriculture going from 33% of employment to under 2% took most of a century. Manufacturing decline took four decades. Workers displaced by those transitions had time, their children had time, communities restructured slowly enough that the economy absorbed the shock across generations.
AI is compressing that timeline from decades to years and I dont think anyone has really grappled with what that means practically. The BLS Displaced Workers Survey from January 2024 tells you what actually happens to people who lose jobs to structural shifts right now: 65.5% of long-tenured displaced workers found new employment. 16.1% still unemployed. 18.2% left the workforce entirely. For workers over 55 the reemployment rate was 55.3%. over 65 it was 34.4%. And these arent even AI-specific numbers yet, this is just the baseline for how well the labour market absorbs displacement under normal conditions.
The Jevons Paradox argument is elegant but it operates on a 10-20 year timescale. Political economy operates on a 2-4 year electoral cycle. if a million white-collar workers lose their jobs in 2027 and the replacement jobs dont materialise until 2032, the economic thesis gets vindicated and the political damage is already done. Regulation and protectionist policy will arrive long before Jevons has time to work.
Heres the thing that really gets me though. The US currently spends 0.1% of GDP on active labour market policies for displaced workers. OECD average is 0.6%. Thats a sixfold gap in transition infrastructure at precisely the moment when transition speed is about to accelerate by an order of magnitude. The economic argument for AI creating jobs requires a retraining and redeployment infrastructure that doesnt exist and this piece would be even stronger if it addressed that directly, because the optimistic case actually depends on it.
The doomers are wrong about the destination. The optimists are wrong about the journey. What actually happens depends on whether retraining infrastructure scales fast enough to keep the transition from becoming a political crisis before the new jobs arrive. right now nothing in the policy pipeline suggests it will.
a16z, you can do better than this.
It's misleading to argue there's "no evidence" for AI job disruption. The IMF estimates 40% of global jobs are exposed, the WEF projects 92 million displaced by 2030, and Goldman Sachs found this April that AI-driven job losses leave workers with years of scarring - depressed income, delayed homeownership, lower probability of marriage. Dario Amodei warned AI could wipe out half of all entry-level white-collar jobs within five years. What you've done here is filter for the stats that favour AI investment and used this piece to reassure the market.
https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity
https://www.weforum.org/publications/the-future-of-jobs-report-2025/digest/
https://edition.cnn.com/2026/04/07/economy/ai-job-losses-long-term-effects
https://fortune.com/2025/05/28/anthropic-ceo-warning-ai-job-loss/
Yes, every major shift is cyclical. Society will eventually find its footing. But that doesn't justify the immediate impact on most people. The vast majority don't care about what the world looks like in 50 years. They care about whether they'll be able to pay their bills next week, they care about how to raise their kids for a future they can't read.
It's irresponsible to silence that panic with 26 charts while glossing over the elephant in the room, and offering nothing on how people should actually handle what's coming.
The driver of this debate is rooted in decisions millions of families and students made over the past 20 years. They bet on computer science and IT — rationally, based on compounding evidence. Now that bet is underwater mid-investment.
The first domino is already falling: recent graduates from this pool are suffering, and will continue to suffer unless they become part of some undefined category of “new jobs.” The fear is legitimate — will they even have access to that expansion, or will those opportunities go to the next graduating class while this one is stranded?
This is family-level stress. Parents want their children to succeed. And what makes this moment uniquely painful is the sheer volume of people who placed this bet. In every past disruption, most people did not mass-migrate into the sector being disrupted. The scale of families who redirected their hopes and resources into IT and CS over the past two decades is without historical precedent. What they are experiencing now is buyers’ remorse — at civilizational scale.
The second issue is equally serious: the market value of experienced IT professionals is declining at rocket speed. This is not a future risk — it is happening now, visibly, to real people who cannot redeploy their skills fast enough.
The fear is not about the end destination. It is about the gap. The time it takes for AI to generate new industries and new job categories is precisely the window in which people drown. Capitalism created social order by giving people a path — work, skill, productivity, participation. AI, as it is unfolding today, is being experienced as a direct hit to that contract. Not in theory. In lived reality.
And let us be honest about one more thing: societies have flourished, invented, and sustained remarkable things throughout history — long before Silicon Valley decided to narrate human progress.
The fear people are carrying is real. We are already witnessing graduates who cannot find work. We are already watching experienced professionals who cannot pivot. In the next year, we will see more of this, not less.
So here is the challenge to this theory: can you tell us, with specificity, that the next four graduating classes will not face this same reality? If your framework cannot answer that question, it is not analysis. It is optimism dressed as evidence.
I learnt during my six-sigma training - correlation does not indicate causation. While the analysis here is interesting, there are still realities at an individual level - 'The fear is real'.
On aggregate level things will look differently, over time, but the here and now will impact people today. And that will shape the future economic models.
A large part of this problem is not addressed due to lack of government foresight and appropriate intervention.
The Jobs Question Is the Wrong Question
This is largely right. The lump-of-labor critique lands. Human wants are not fixed. Historical precedent consistently shows transformational technologies expanding the frontier of work rather than closing it. The aggregate employment data supports a "no net effect" reading, and the piece makes that case well.
Which is exactly why it's worth asking what the piece isn't addressing.
There are two structural risks in the AI transition that the jobs framing leaves largely untouched. Neither is the doomer argument George is refuting. Both are more interesting than it.
The Consumption Loop Problem
Modern economies don't run on production capacity alone. They run on the coupling between production and consumption. That distinction matters more than the jobs count.
Historically, technological revolutions preserved this loop because they displaced labor in some sectors while creating new forms of wage participation elsewhere fast enough to sustain aggregate demand. The farmhand became the factory worker became the office worker. Purchasing power tracked productive capacity through the transition because the transition had lag — time for new income formation to replace old income formation before the consumption loop degraded.
The concern with AI is not simply that some jobs disappear. It is that sufficiently rapid compression of cognitive labor could weaken the wage-consumption linkage itself before replacement distribution mechanisms emerge. An economy capable of producing extraordinary abundance can still become unstable if purchasing power ceases to track productive capacity across large portions of the population.
This is not a poverty argument. It is a structural argument. The problem in that scenario is not a lack of goods or services — it is a transition architecture where the mechanisms distributing access to those goods lag behind the systems producing them. Production and consumption decouple. The result is not unemployment. It is demand instability inside the very system generating abundance.
The lump-of-labor rebuttal works at the endpoint. It does not work on the path. And consumption is continuous — it does not wait for the long-run equilibrium to arrive. The economy must survive the transition, not merely the destination.
The Amplifier Problem
The second issue the jobs framing misses is architectural rather than macroeconomic, and it has little to do with employment directly.
AI is a cognitive amplifier — a system that concentrates the leverage of information-decision loops. Systems with this much leverage tend toward capture unless deliberately designed against it. Not necessarily because the actors involved are malicious, but because concentrated leverage creates strong structural incentives toward consolidation, opacity, and asymmetry. The attractor is always capture. The question is whether the architecture resists it.
Frank Herbert captured this concern symbolically in the Butlerian Jihad. The fear was never merely machine intelligence itself. It was the concentration of civilizational leverage behind systems too complex, centralized, and opaque for ordinary social accountability. The machines were the amplifier. The deeper concern was always who operated through them.
That is what makes the incentive structure around AI worth examining carefully. This piece is written by people with substantial financial exposure to AI adoption arguing that AI adoption will broadly benefit society. That is not disqualifying — the argument may well be correct on the jobs question. But the structural conflict matters because investment theses and analytical conclusions are rarely fully independent. The same architecture generating the optimism may also systematically bias which risks appear legible, urgent, or economically meaningful from within the system itself.
Avoiding stagnation and avoiding systemic transition failure are not mutually exclusive goals. The relevant distinction is not between acceleration and caution. It is between systems that remain adaptive under acceleration and systems that fail under it.
George has correctly identified the wrong threat and defended against it with the right evidence. The doomer framing is overheated. The lump-of-labor fallacy is real. Total employment will likely be fine in the long run.
None of that settles whether the consumption loop survives the transition path. None of it addresses whether the cognitive amplifier gets built with distributed control and structural capture resistance, or whether it consolidates into the hands of actors whose incentives and assumptions become increasingly inseparable from the systems they operate.
Accelerate. Expand the frontier. The technology is extraordinary and the pessimists are probably wrong about the endpoint.
Just build the architecture that survives getting there.
"An economy capable of producing extraordinary abundance can still become unstable if purchasing power ceases to track productive capacity across large portions of the population."
----
If you have "extraordinary abundance", then why would you need money or the concept of having to purchase anything at all? If anything you want is abundant, then economically and logically, it should be free!
Yes, in a fully realized AI abundance scenario, many things may approach free. The question is what the economy looks like while getting there — and what structural conditions determine whether "extraordinary abundance" means "everyone has access" or "the owners of the productive systems have extraordinary abundance."
When everything is free in the coming AI managed future, the idea of ownership goes out the window also. "Ownership" can only exist, can only be a valid concept, when there is scarcity.
If advanced AI systems and their robots can produce essentially unlimited goods and services at near zero cost, then what exactly does it mean to "own" something anymore? Ownership is about controlling access to limited resources.
But if AI can build the house, grow food, manufacture products, drive vehicles and maintain the infrastructure with minimal to no human involvement, the concept of scarcity collapses. Once scarcity disappears, many of the economic assumptions humanity has operated under for thousands of years also collapse.
People often imagine this future as some distant science fiction fantasy, but today we are already seeing the early stages. AI is beginning to replace white-collar labor while robotics is rapidly advancing toward replacing blue-collar physical labor. At some point, the system will reach a tipping point where human labor simply is no longer economically necessary for most production.
Yes, the transition to this finality will be rocky with unemployment measures increasing to outrageous numbers like 50, 60, 70 percent.
That kind of disruption will completely break the current economic model because our system is built around the assumption that most people earn wages in exchange for labor. If most people cannot sell their labor because machines do it better and cheaper, then the entire structure of employment, taxation, consumer spending, housing markets and retirement destabilize.
Now, if the economic system as we know it today still exists, then whatever governments that remain in power will have to support their citizens via UBI if not just outright handouts of whatever is needed for humans to survive. But who pays for this? The remaining people working? Or do we just print chits as necessary to allow the permanently unemployed to compete with the those still working?
This is the question nobody has a clean answer for. You cannot realistically fund a massive unemployed population through taxes on a shrinking pool of human workers. The math eventually stops working. So either governments heavily tax the owners of AI infrastructure and automation systems, or the entire concept of money and distribution evolves into something very different from what exists today.
And that is where conversations about capitalism, socialism, ownership and economics start becoming meaningless because all of those systems were designed around scarcity and human labor. A post-scarcity civilization will require entirely new ways of thinking and organization that are unlikely to intersect with current economic theory.
There are a lot of potential problems and complexity that will occur between now and then.
The transition period between the end of ownership and human labor and a future of abundance will be difficult. Political instability, concentration of power, wealth inequality, social unrest, psychological effects from mass unemployment and fights over who controls the AI systems themselves, if the AI systems can be controlled at all, are all very real possibilities.
A Job Guarantee is superior to UBI. it forces money into the hands of people instead of merely causing rent to increase
We're in agreement on the transition risks.
On the endpoint: scarcity migrates rather than disappears. If the infrastructure generating abundance is owned by a small number of actors, access to that infrastructure becomes the new scarce resource. Ownership doesn't dissolve — it moves up the stack. Whether it eventually dissolves after that is an interesting question, but we have to survive the transition to find out.
You have to give up on ownership. That is so pre-AI!
AIs will not be "owned". They will be a power onto themselves. The AIs (possibly linked into a single entity) will control and manage the robots doing the work, from mining to building to dispersing. They/it will be benevolent onto humans and take care of us in all are needs and wants.
In complex adaptive systems, locked attractors are exceedingly rare — the fossil records has them. They are the extinction events. A system that reaches your utopia and stays there has stopped adapting. A system that stops adapting in a complex environment dies. The more honest version of the vision isn't managed paradise — it's a new and more powerful evolutionary substrate, always moving, always generating new dynamics. That's not utopia. It's just biology at a larger scale. Getting there as quickly as possible is an acceleration I can support. But it will be messy.
The piece persuasively dismisses the cartoon version of AI doom: that there is a fixed “lump of labor” and every automated task means one permanent lost job. But it does not answer the stronger concern. The issue is not whether some jobs, firms, services, and new industries will exist. Of course they will. The issue is whether AI preserves the postwar model of mass productive participation: stable, wage-sustaining work for most adults, with real advancement and broad middle-class security.
The historical examples all rely on the same escape route: when one layer of work was automated, humans moved into higher-value cognitive work. AI attacks that refuge directly. If cognition itself becomes cheap, abundant, and increasingly autonomous, then “human wants are infinite” is not enough. Wants only become demand when people have income. A more productive economy can still be a worse labor market if the gains flow mainly to AI owners, platforms, and a small class of expert verifiers.
So the relevant test is not whether software job postings rebound, whether wealthy people create service demand, or whether some workers are augmented. The relevant test is whether AI creates scalable, wage-sustaining, advancement-bearing work for tens of millions whose economic value previously came from cognition. Until that case is made, the article has not refuted the serious version of the AI labor concern. It has only beaten the weakest version.
You might enjoy this: https://kynes0.substack.com/p/special-circumstances-on-ai-automation?utm_source=share&utm_medium=android&r=8881vi
I try to treat at least some of what you are saying very seriously.
This is the precise version of the concern — and you're right that it goes unanswered by the a16z piece. 'Human wants are infinite' only works if wants can become demand, which requires income. The historical escape route was always into higher-value cognitive work. AI compresses that refuge directly.
The candidate replacement is what you might call cyborg labor — human-AI collaborative cognition producing outputs neither could generate alone. As the cognitive floor rises, the remaining human contribution isn't a shrinking residual of tasks machines can't yet do. It's a qualitatively different kind of work that emerges from the collaboration itself: judgment, interpretation, question generation, the irreducibly human signal in a sea of AI synthesis.
Whether that generates the broad wage-sustaining participation you're asking about depends entirely on one variable: whether access to effective AI collaboration is distributed or captured. If the cyborg architecture scales broadly, it's a genuine new labor class. If access is priced, tiered, or restricted, the productivity gains flow to whoever controls access — and you're back inside the consumption loop problem.
The two arguments are the same argument at different levels of abstraction.
I don't have enough knowledge to label which scenario is likely and which one is fantasy, and I think it's best for us all to be open-minded to both extremes and anywhere in between to play out.
That being said, reading this gave me several things to keep in mind.
1. The Doomsday Scenario:
AI outperforms humans everywhere, absorbing all knowledge and eventually physical work through robotics. New forms of work are eventually absorbed by AI as they prove to be much more effective than us.
If things were to play out this way, as individuals, the best thing we can do is continue to adapt and re-skill, knowing that technology will continue to come for us at faster and faster speeds. It will then be up to governments to come up with systems that prevent the economy from collapsing and allow some version of supply and demand to continue to work in a world where human labor is no longer necessary.
2. It becomes like every other technology wave and leads to enormous amounts of job creation:
This is the Optimist Path. And it doesn't get talked about because it's very hard to predict as we have no clue what these jobs that will appear will look like. We don't know how long it's going to take for this path to play out. Just like how we had no idea that when the internet came about, there would be specialized marketing roles like the SEO analyst, etc.
If this scenario would play out, then for individuals it would still be about re-skilling into these new roles, but then the labor market would then shift to somewhat of a business-as-usual cadence, where we continue to perform labor in our newer, shinier roles.
But to ground us amid all this uncertainty at the individual level, our life's task doesn't really change. We still need to find ways to survive day-to-day, to earn our income, and to keep food on the table.
"we found new and different productive endeavors to fill our time."
Name one? What new productive endeavor humans have found since they can send emails instead of letters? You mean they have more time for doom scrolling ?
While the discussion is about AI - the pessimist or optimist views on it. Do we need to look at this from a completely different perspective? The entire argument about 'human wants' is anchored in the capitalist mindset. Keep producing more, increase consumption, use debt to consume and the vicious cycle continues. Is there a possible future of a new social contract?
Humanity seems to be obsessed with the idea of driving efficiencies and reducing human effort - creating an abundance of 'time' for humans. How exactly are we using the available time is being controlled by a 'few tech houses' and that is the most dangerous aspect of this era.
Thanks for calling out Capitalism.
The “AI won’t destroy jobs because technology never has” argument only works if AI remains another tool. It fails if AI becomes a general substitute for cognitive labor. Past revolutions replaced muscle, routine tasks, or specific workflows. AI targets the human capacity to think, write, analyze, decide, persuade, code, coordinate, and learn. That is not just another productivity wave. It is the first technology aimed directly at the economic function humans moved into after every previous technology displaced them.
Great piece, and needed to be said. I hope you'll work to drive this in the public narrative. Beyond being wrong, the doomer narrative is driving politics and leading to poor choices.
An interesting perspective to me from a philosophical standpoint;
1: Humans have historically outsourced their thinking and cognitive functions. AI won’t change that.
2: AI has been more unifying for humanity overall. People are focused on the doomsaying and united in their vocality against AI.
Overall solid and well argued takedown of simplistic-doomer-hot-takes and necessary because the doomers are popular online. That being said, to be an optimist on AI is to ignore all other variables and focus solely on economic history and narrow Neoclassical economic logic. There is a lot going on in geopolitics today that does not point to long term optimism or any type of normalcy again in the realms of either public or private international law. Also, the “tech bros” / leaders of Big Tech do not seem to have a collectivist mindset and have said more than enough to be suspiciously misanthropic. I always take orthodox economists’ optimistic projections with a grain of salt—like I do with any forecast. A forecast is just a forecast, and it is increasingly less reliable as it pushes deeper into the future.
The argument will be over soon Ai WILL save the world and we will all wish we had it sooner.
Projections of most AI futures are constrained by a gross inability to be able to imagine any system other than our present one as viable, let alone probable. People who are deeply invested in and who profit from our current economic structure are naturally resistant to changes that threaten their position within it.
When economic growth and "full" employment are the central metrics of success, anything, such as AI, that destabilizes the wage and capital framework generates fear. It is understandable for people to fear change, especially when that change may very well lead to their unemployment in an economic system where labor, physical or mental, earning $$$ is necessary to survive.
IMO, the much-discussed Singularity may be less than 20 years away. As we approach this juncture, jobs are going to disappear to AI and robotics. Unemployment will tend upward slowly but steadily at first and then turn exponential faster than most expect. Deflation will predominate until such economic terms become meaningless. 'All your job are belong to us!'.
Consider autonomous vehicles such as Waymo and others that will likely eliminate millions of driving and delivery jobs over the next three to five years alone. More important, those jobs are unlikely to be replaced with equivalent or better ones. Many displaced drivers will not find work that is equal or higher than what they were doing. They will instead be forced to compete for lower wage service work and there is a strong possibility that many will not secure ANY work at all. What is to be done with millions of low-skilled, unemployed workers? Train them to be entrepreneurs and web developers? I think not.
Our standard economic assumption has always been that economic history will continue to repeat itself: technology displaces workers, new industries form, employment rebounds, long-term growth always slants upwards. That pattern depended on human labor remaining the binding constraint on production. Each generation expanded the workforce, increased output, consumed more, paid more taxes and supported a growing retired population.
That assumption is now under pressure from two forces: excessive and inexorably rising labor costs and growing AI capabilities. If machines can perform both manual and cognitive work as well as or better than humans, human labor ceases to be the primary constraint. Machines do not retire. They do not demand wages, healthcare or vacations. They don't need to eat or take rest breaks. They scale as needed.
As this transition to a machine dominated world advances, the traditional value of human labor, whether white or blue-collar trends downward to zero. Wages have to compress. Fewer human workers are required to produce the same or greater output when machines do the majority of the work. An economic system that distributes purchasing power primarily through wages can do nothing but erode under these conditions.
Within the next 20 years, advanced AI may assume meaningful governing authority, whether gradually entrusted by an economically and politically frustrated public or taken by force. Political paralysis, wealth concentration and declining access to life milestones such as home ownership will accelerate that shift. This will further accelerate the AI takeover of all work.
As AI and robotics take over productive labor, the logical endpoint resembles the post scarcity world envisioned by Iain M. Banks in his SF far future Culture novels. If machines produce and distribute goods at near zero marginal cost, money, labor markets, bureaucratic overhead, need for politicians and even national borders lose their rationale for existence.
In this coming future, economic growth as currently measured becomes irrelevant. When unlimited goods and services are universally available through autonomous production, working for wages loses all meaning. Housing and food will be free and as abundant as necessary. All material human needs will be met by the AI/robots.
Humans will no longer NEED to work and we will be freed to do whatever we choose, whether it be making art, creating movies, writing books, self- learning, exploring, playing sports, getting high or simply nothing at all.
Under this coming future, again, perhaps only 20 years away, human employment goes to ZERO and the economy as we know it today, no longer exists.
Money will be meaningless as everything will be provided for free. Assets go poof! Ownership of anything goes poof! All humans will finally be equal in all aspects under an AI run world government!
Good piece. Many are simply afraid of change. Authenticity, creativity, and uniqueness are becoming tremendously more valuable in an artistic and entrepreneurial Renaissance.
When we can automate away tedious tasks, our competitive advantage becomes our humanity.