III. Work’s Last Stand

When labor turns optional, does society thrive—or fracture? The AGI transition will fundamentally rewrite work and capitalism. Unlike past automation, AGI targets cognitive labor directly. Expect AI assistants (2025-30), then broad automation (2030-45), leading to a post-employment economy (2045+). This existential shift redefines the social contract. The core questions are which jobs endure and how we distribute the gains.

   [The work forecast](#the-work-forecast)

   [Faster than ever](#faster-than-ever)

   [The third path](#the-third-path)

   [Why some jobs endure](#why-some-jobs-endure)

   [Hard to replace jobs](#hard-to-replace-jobs)

   [The way ahead](#the-way-ahead)

   [Next 5 years (2025-2030)](#appendix-1.-next-5-years-2025-2030)

   [Medium term (5-20 years: 2030-2045)](#appendix-2.-medium-term-5-20-years-2030-2045)

   [Long term (20+ years, 2045 and beyond)](#appendix-3.-long-term-20-years-2045-and-beyond)

   [Limits to replacement](#appendix-1.-limits-to-replacement)

   [Hard to replace jobs](#appendix-2.-hard-to-replace-jobs)

The greatest danger in times of turbulence is not the turbulence; it is to act with yesterday’s logic.

— Peter F. Drucker

If you’re enjoying this deep dive and want to keep up to date with my future writing projects and ideas, subscribe here.

The new economic order

What happens when the most transformative technology in human history arrives not over generations, but all at once? That's the question we face with AGI.

Previous technological waves—agriculture, industrialization, computing—gave us time to adapt. AGI won't. It will compress centuries of change into decades, maybe even years.

In this essay, we'll focus on how AGI will transform work and capitalism.

Why focus on these areas? For one thing, money matters. Globally, higher income correlates with longer lives¹, better health, and greater happiness. Also, work is where AI will first collide with human systems—where algorithms meet payrolls, labor regulations, and status hierarchies.

When productivity transforms, it creates ripple effects. These cascade through price structures, tax bases, and even geopolitical power balances. And honestly, I find it fascinating to think about how power, creativity, and rewards might redistribute when intelligence—previously our scarcest resource—becomes abundant.

But before we dive in, a note on uncertainty. Projecting second- and third-order effects is speculation, not science. Mistakes are guaranteed. But, as they say, it's better to be roughly right than precisely wrong.

Employment's next era

Something unprecedented is happening. For the first time in history, we're approaching a point where technology might not just change work but potentially make much of it unnecessary. And not in some distant future—it's beginning to happen now.

Every previous technological revolution followed a reassuring pattern: yes, some jobs disappeared, but more new ones were created. The tractor replaced farm workers, but factories emerged. Computers eliminated clerical jobs, but created an entire tech industry. Economists have long pointed to this pattern as proof that technology ultimately benefits workers.

But AGI breaks this pattern.

Why? Because previous technologies replaced human physical capabilities. AGI replaces human thinking. And thinking is what we retreated to when our physical labor was automated. If both mind and muscle can be replicated by machines, what's uniquely left for humans?

The signs are already visible. Labor conditions for new graduates have deteriorated sharply in recent months, with unemployment reaching an alarming 5.8 percent. Elite M.B.A. graduates are struggling to secure positions. Simultaneously, law school applications are surging—an ominous parallel to the great financial crisis when young people retreated to graduate education as shelter from economic turmoil.

There are several factors contributing to this backdrop, including an uncertain economy and higher-interest rate environment. But it's hard to deny that artificial intelligence might be starting to play a role.

An increasing number of public company CEO's are publishing staff memos like the below.

Competitive forces dictate that this will happen broadly at an increasing rate—potentially with little regard for junior workers.

This creates a complex societal question: who's responsible when labor becomes optional? Who is going to take responsibility for labor market displacement? We don't yet have a good answer to this question. And I'm sure it leaves a lot of people feeling like this:

View on X →

The tension here goes beyond economics. It touches something about human identity. We've built our entire social structure around work—not just for income, but for purpose, status, and meaning. What happens when that cornerstone starts to crumble?

Part of the problem is that different people see different parts of the puzzle. Technologists understand what AI can do but not necessarily what that means for society. Social scientists grasp the human impact but often don't fully comprehend the technology's trajectory.

What makes AI different from previous innovations is its target: cognition. Steam power and electricity mechanized physical work. Early computing automated narrow information processing. But AGI challenges what we've long considered uniquely and permanently human—our ability to think, reason, and create. Not because machines will think exactly like humans do, but because they'll increasingly perform the economic functions² we previously reserved for human thought.

We face not just an economic reckoning but a philosophical one. Much resistance to AI stems from a deeply embedded belief that difficulty signals virtue—an industrial-era value system where suffering equals merit and effort equals worth. We've built our status hierarchies around rewarding the inefficiencies that technology eliminates.

The anxiety surrounding automation is about worth and meaning. When traditional metrics of value—hours worked, credentials earned, specialized knowledge acquired—collapse under AI's efficiency, what happens to human dignity?

If AGI makes most cognitive labor effortless, we'll need to decouple dignity from drudgery, just as we once separated human value from physical strength. This is a psychological problem: how do we create new foundations for meaning that don't depend on artificial scarcity or performative struggle?

I'm not writing this to induce resignation but to inspire preparation. The path ahead contains both threat and extraordinary possibility. The outcome depends not on technological determinism but on how wisely we respond to what may be humanity's last invention.

What's at stake is the future of what it means to be human in a world where human labor—both physical and mental—is no longer the primary engine of production. That's a future we need to start planning for now.

What follows is the scenario I see most likely to unfold, based on a deep dive and personal opinion. In many ways, I hope I'm wrong, because the world is not ready for this timeline.

Faster than ever

Whenever a major new technology appears, it follows a predictable pattern³: the world gets excited, then not much seems to happen for a while, and then gradually, over decades, the technology transforms society. Economic historian Carlota Perez has documented this pattern meticulously. Steam power, electricity, and the internet all followed this trajectory—typically taking 20-30 years before they showed up as measurable productivity gains.

The delay isn’t because the tech isn’t good enough. It’s because the world around it isn’t ready. You need new standards, infrastructure, business models, and skills.

If this pattern holds, artificial intelligence should diffuse gradually through the economy. We'd expect to see modest initial effects—some routine cognitive tasks⁴ would be automated while new roles orchestrating AI systems would emerge. Productivity would rise slowly at first, with dramatic acceleration only when Companies restructure their operations around AI-native processes.

But I'm increasingly convinced that AI will break this pattern.

Unlike previous transformative technologies, AI may compress decades of diffusion into years due to structural advantages unique to our digital era.

Previous technological waves required building physical infrastructure from scratch—electricity needed generating plants and transmission lines, the internet demanded fiber optic cables and routing hardware. AI, by contrast, rides on existing digital infrastructure, enabling what we might call "one-switch rollout" potential. Tech giants already controlling billions of users' digital experiences can transition hundreds of millions to AI-augmented workflows overnight through mere software updates.

Modern software architecture further accelerates this process. AI capabilities deploy through standardized APIs, allowing any application to AI with minimal friction. Open-source models are democratizing access to capabilities once confined to elite private companies, paralleling Linux's impact but with potentially greater velocity across economic sectors.

Most significantly, AI possesses a unique capacity to facilitate its own integration—generating implementation code, testing protocols, and documentation, while providing personalized instruction adapted to each user's needs. These mechanisms function as an integrated system with compounding effects: AI trains its own users, orchestrates its own implementation, and continuously enhances its capabilities.

This feedback loop suggests radical compression of adoption timelines. What historically required thirty years might now take five. If AI diffuses at even half this potential speed, our institutions will face transformative pressure before developing adaptive capacity.

The assumption that technological waves take decades to play out has led many to believe we have time to prepare. But the digital world AI enters operates under different rules. I'm not suggesting the economy will transform overnight, but the comfortable assumption that we have decades to adapt may be dangerously wrong. The acceleration mechanisms are already in place, and they're unlike anything we've seen before.

AI is unlike any other general-purpose technology riding the old rails. It’s laying new track as it moves, training its own users, and even updating itself on the fly. That kind of compounding feedback loop could shrink a 30-year diffusion timeline down to something closer to 5 or 10 years.

The work forecast

Throughout history, technology has changed the nature of work gradually enough for society to adapt. New institutions emerged as old ones faded away. People had time to learn new skills as their old ones became obsolete. We developed new sources of meaning as traditional ones lost relevance.

As we just explored, AI is unlikely to be so accommodating.

Instead of gradual change, we're facing a compression of history—changes that previously took generations will happen in just years or decades. We need to prepare not just for immediate disruptions but for the cascading changes that will reshape human labor.

I've been thinking about how this might unfold across three distinct phases.

Next 5 Years (2025–2030):

In this first phase, AI functions primarily as a collaborator rather than a replacement. These systems amplify our cognitive abilities—handling routine tasks while humans maintain control over orchestration, judgment, and creative synthesis.

White-collar professions won't disappear but will reorganize internally. Legal associates will delegate document review to algorithms but maintain client relationships. Radiologists will use AI for diagnostic assistance but provide the integrative interpretation that connects medical findings to the full patient context. Educators will automate assessment but deepen their focus on mentorship.

Job losses will be concentrated in predictable areas where routine cognitive tasks dominate. Some of these losses will be offset by new roles in AI development, oversight, and implementation.

The most significant change won't be in unemployment rates but in productivity disparity. Those who effectively collaborate with AI will outperform those who don't. This represents a revaluation of human skills. Metacognitive capabilities, ethical judgment, and interpersonal intelligence will appreciate in value, while routine analytical work will depreciate.

Governments will respond reactively rather than transformatively during this period. They'll emphasize retraining programs, educational realignment, and incentives for human-AI collaboration, but won't reconsider the nature of work. This reflects our tendency to imagine the future as a modest extension of the present rather than something qualitatively different.

Medium Term (2030–2045):

Around 2030, with the emergence of AGI, everything changes. Automation expands beyond narrow domains to encompass nearly every field of economically valuable cognitive labor.

Employment displacement accelerates. Projections suggest between 10% and 30% of the global workforce may become obsolete in their current roles. New industries will emerge from these technological capabilities, but they’re unlikely to create enough jobs to replace those lost—challenging the historical pattern where creative destruction ultimately generated more opportunities than it eliminated.

The remaining areas of human advantage—roles requiring contextual awareness, emotional intelligence, moral reasoning, and creative originality—will temporarily become more valuable before facing incremental encroachment as AI capabilities continue advancing.

Income polarization will intensify as ownership of autonomous productive capacity concentrates, while labor's bargaining position erodes through technological substitution. Without robust safety nets, wealth transfers, or restructuring of ownership, social cohesion will face mounting pressure as economic displacement translates into political alienation.

This period represents the fulcrum upon which the long-term social contract balances. Societies will either develop new mechanisms to distribute abundance divorced from traditional labor contribution or face escalating instability as technological efficiency collides with distribution systems designed for an era when human labor was central to production.

The window for adaptive policy innovation narrows as technological advancement accelerates. We'll need foresight rather than reaction if we want harmony between human thriving and technological advancement.x

Long Term (2045+):

Conventional employment—exchanging human cognitive or physical labor for compensation—will cease to serve as the primary mechanism for resource distribution or as the central organizing principle of adult life. Human activity will undergo redefinition, shifting toward domains chosen for intrinsic reward rather than economic necessity—creative expression, relationship cultivation, philosophical exploration, and forms of contribution valued for their uniquely human character rather than their productive efficiency.

This transition will necessitate restructuring of economic distribution mechanisms. Universal basic income, substantial taxation of automated production, or novel ownership structures for artificial productivity will become existential necessities if prosperity is to be broadly shared rather than concentrated among the vanishingly small fraction of humanity controlling autonomous productive capacity.

The definition of work will undergo philosophical change—no longer signifying what one must do to survive, but rather what one chooses to do to thrive. The potential outcomes span extremes—from dystopian concentration of power enabled by production systems that require minimal human input, to unparalleled flourishing where liberation from necessity enables expanded human potential.

The determining factor lies in the wisdom with which societies navigate this transition—whether they cling to distribution paradigms designed for an era of labor scarcity or develop new frameworks appropriate to an age of abundance generated through autonomous production.

The end of an era

What becomes clear across all three horizons is that artificial intelligence challenges the relationship between labor and livelihood that has defined human experience since the dawn of civilization.

The mathematical certainty that has reassured economists throughout previous technological transitions—that productivity gains would ultimately generate more opportunities than they eliminated—faces challenge when the technology in question replicates not specific human capabilities but generalizable intelligence.

In navigating this change, we confront a philosophical reimagining of humanity's relationship with work, worth, and one another.

This is no longer speculation about a distant future. The early signs are already visible. How quickly it happens and whether we'll be prepared when it does are the core questions, rather than whether it will happen.

Work after AGI

The third path

Work defines not just how we survive but who we are. It gives us purpose, status, identity, and a sense of belonging. As we approach AGI, this ancient relationship will come under the microscope to a degree we’ve not seen before.

Most discussions about the future of work fall into two camps. On one side, you have people who assume AGI will automate everything, leaving humans with nothing to do. On the other, you have optimists who think AI will just "augment" work, like a fancy Excel macro, without ever really replacing us. Both miss the more nuanced reality⁵ that's likely to emerge.

Even in a world where artificial intelligence achieves comprehensive superiority across cognitive and physical domains, certain kinds of human labor will persist. Not because machines can't perform these functions, but because specific dimensions of human activity retain distinctive value beyond pure efficiency.

Drawing from economists like Daniel Susskind and Deric Cheng, we can identify three principles that will preserve certain domains of human work in the post-AGI landscape. Think of it as a field guide to the last islands of human labor in a post-AGI economy.

General Equilibrium Limits

Even if artificial intelligence achieves absolute advantage across all domains, the principle of comparative advantage⁶ ensures humans would retain certain roles—at least for a while. David Ricardo articulated this economic reality back in 1817—optimal efficiency occurs when each entity focuses on activities where it has the lowest opportunity cost.

In practical terms, this means that in a world of hyper-competent artificial intelligence, human labor would naturally flow toward domains where the relative efficiency gap is narrowest. Or where human performance, while technically inferior, remains "good enough" relative to the cost of full automation. It’s cheaper and therefore doesn’t get automated.

Preference Limits

More significant than the economic argument is human preference⁷. Some domains will persist not because machines can't perform them, but because the human element constitutes an essential aspect of their value.

Consider creative expression: we often value works specifically because humans created them. Or competitive achievement, where meaning derives from human limitation and striving. Or empathic connection, where authenticity matters more than perfection.

A sonata loses something essential when composed by an algorithm rather than human imagination. A competitive achievement means less without the narrative of human struggle behind it. Therapy delivered without genuine empathy becomes mere information transmission rather than healing.

Moral Limits

Certain domains will remain human because we collectively determine they should—recognizing that some decisions require not only intelligence but moral legitimacy.

The administration of justice, the governance of institutions, the alignment of technological systems with human values—these domains demand not just competence but rightful authority. Even as narrow systems advise judges, broader moral questions about punishment, rehabilitation, and justice remain intrinsically human.

As AGI emerges, humans may occupy a meta-role: ensuring alignment between autonomous systems and evolving human values.

These domains of preserved human labor shouldn't be understood as mere economic residuals—the jobs machines happen to leave behind. Rather, they represent expressions of what we determine to be essentially human. They're not just what we can still do in a world of hypercompetent machines, but what we choose to keep as uniquely ours—preserving not just employment but meaning, dignity, and purpose.

For an extended discussion of the limits to automation, see Limits to replacement.

Hard to replace jobs

Now that we have a framework, we can look at specific jobs that will survive — at least for a while.

Before diving in, we should acknowledge three important points. First, these areas won't remain static. Even as certain roles persist, they'll transform beyond current recognition—preserving their essence while changing their form. Second, most current jobs will likely yield to automation eventually, making these exceptions increasingly precious forms of human contribution. Third, there's a crucial distinction between technical feasibility and economic rationality—just because AGI can do something doesn't necessarily make human alternatives obsolete in all contexts.

I'm not exploring these domains to offer false comfort. I want to show where human labor might retain distinctive value. I'm explicitly excluding jobs preserved through artificial scarcity or temporary cost advantage—focusing instead on work that remains human because something essential would be lost in full automation.

What follows is a field guide to the last islands of human work.

1 Legitimacy & Ritual Authority

Heads of state, judges, and military commanders make decisions requiring not just intelligence but legitimate moral agency. People trust judgments from these roles because humans with personal accountability render them.

While AI will draft the briefs, analyze precedents, and model scenarios, the final authority will likely remain human—concentrated in smaller cohorts serving as the interface between algorithmic recommendation and social implementation.

2 Empathy-Driven Care & Guidance

Nurses, therapists, and teachers occupy a domain where embodied warmth and improvisational understanding matter deeply, particularly in contexts of vulnerability. The distinction persists not because machines cannot diagnose or develop treatment plans—they will—but because genuine human presence constitutes part of the healing or developmental process.

As administrative aspects yield to automation, human practitioners will focus more on relational dimensions that algorithms cannot authentically replicate.

3 Artisan Craft & Aesthetic Expression

Violin makers, bespoke tailors, and restoration specialists preserve tradition through work where human limitation and imperfection paradoxically create value rather than diminish it. The status of "hand-made" objects may initially appear arbitrary, yet it reflects a genuine appreciation for human effort and historical continuity embedded within physical artifacts.

Think about why people value artisanal products. It's not just about the physical object but the narrative of its creation—the years of apprenticeship, the tradition handed down through generations, the minute imperfections that make each piece unique. While AI will assist with design optimization, artisans will thrive in markets that value the authenticity of human craftsmanship.

4 High-Dexterity Field Operations

Plumbers, emergency medical technicians, and specialists in hazardous environments navigate unpredictable physical contexts where adaptability to chaotic conditions remains challenging for robotic systems. Though drones and specialized robots will assist with standardized aspects of these roles, human leadership will likely persist in coordinating responses to novel situations where embodied intelligence and contextual awareness prove decisive.

5 Experience & Cultural Mediation

Sommeliers, museum curators, and cultural commentators translate sensory and cultural nuance in ways that transcend mere information transfer. These roles persist because taste, ritual, and cultural appreciation remain intrinsically human experiences—valued for the shared human context that gives them meaning.

What makes a great sommelier not only encyclopedic knowledge of wine regions and vintages—something AI could easily master—but the ability to connect with the diner, gauge their preferences (often unstated), and create a memorable experience around the selection and enjoyment of wine. It's about cultural translation and shared experience, not just information delivery.

6 Human Performance & Narrative

Olympic athletes, stage performers, and chess grandmasters engage in activities where human achievement and limitation constitute the very source of their value. Audiences engage with these performances not because they represent optimal execution in absolute terms, but because they embody human striving, narrative, and the dramatic tension of finite beings pursuing excellence.

We've already seen this principle at work. Chess engines surpassed human grandmasters decades ago, yet professional chess hasn't disappeared—it's thrived. People watch Magnus Carlsen not because he's the strongest chess entity on the planet (he isn't), but because his human struggle against limitation creates drama and inspiration that a superior algorithm cannot.

7 AI-Systems Orchestration

Product managers, policy specialists, and human-in-the-loop supervisors occupy the critical interface between human intention and machine implementation. This meta-role—translating messy human goals into structured machine instructions—requires both technical fluency and cross-domain understanding that pure specialization lacks.

I suspect this category will expand. As AI systems grow more powerful, the people who can effectively direct these capabilities toward meaningful human objectives will become valuable. It's not unlike how programming evolved—as we built more powerful abstractions, the emphasis shifted from knowing low-level implementation details to understanding how to structure problems for computation.

8 Frontier Discovery & Problem Framing

Deep-tech researchers, theoretical scientists, and investigative journalists operate at the boundaries of knowledge where asking the right questions often proves more valuable than processing existing information. The abductive leaps required to formulate novel hypotheses and recognize emergent patterns in seemingly unrelated phenomena remain distinctively human capacities.

Though AI will accelerate research throughput, human scientists will continue steering inquiry toward the most promising and consequential directions. The essence of scientific discovery is imagination—the ability to ask "What if?" in ways that transcend existing paradigms.

9 High-Stakes Human Persuasion

Negotiators, dealmakers, and mediators navigate complex human motivations and unspoken intentions in contexts where embodied credibility and intuitive understanding of hidden agendas remain decisive.

These roles persist not because machines cannot optimize negotiation strategies, but because human counterparties may simply refuse to engage with non-human entities in domains where trust, relatability, and perceived ethical agency remain prerequisites for participation.

These domains collectively represent expressions of enduring human qualities that retain value even in a world of AGI. They remind us that while machines may match or exceed human capability in isolated dimensions, the integrated expression of human intelligence, creativity, empathy, and moral agency continues to create forms of value that transcend pure information processing.

What's most interesting about these enduring domains is what they reveal about human nature. The work that resists automation clusters around our deepest values, our need for connection, our appreciation of narrative and struggle, and our desire for moral agency in decisions that shape our lives.

The future of human work is unlikely to be about competing with machines on their terms—processing power, memory, or pattern recognition. It will be about leaning into the dimensions of human experience that we uniquely value because of their embodied, fallible, and morally weighted nature. In a world of AGI, human work may paradoxically become more human than ever.

The way ahead

The good news is, even as AGI reshapes the world, there will still be work left for humans — at least for a while. The future of human labor won't be about competing with machines at their own game. It’ll be about leaning into what makes us different: trust, creativity, judgment, presence, and the ability to navigate messy, unpredictable worlds.

But it’s important to be clear-eyed about the limits of these limits.

Over time, the edges where humans have an advantage will keep shrinking. Economic pressures will slowly eat away at general equilibrium advantages. Preferences for human touch and performance could shift as people grow accustomed to AI-created experiences. Even moral guardrails — the places where we insist on human oversight — could weaken if humans are seen as ceremonial rubber stamps rather than real decision-makers.

The framework we’ve mapped here is a guide for the next few decades — a way to find footing while the ground is still moving under us. Our real challenge is finding new ground to stand on, or ensuring that everyone has a lifeboat.

Coming up next: In an economy flooded with intelligence, what still commands a premium?

Continue reading →

Want me to send you new ideas?

If you’re enjoying this deep dive and want to keep up to date with my future writing projects and ideas, subscribe here:

Subscribe to future updates →

Appendix: Work after AGI

The most likely path for employment in a world of imminent AGI is one of accelerating disruption and deep transformation. In the next few years, today's AI will augment and partially automate a lot of tasks. We’ll see noticeable job churn — roles shifting and changing — but not mass unemployment yet. Over the medium term (5–20 years), as AGI arrives and spreads, automation will reach almost every industry. That will mean real displacement — a lot of people will lose their current jobs — even as new industries and new roles start to appear.

In the long run (20+ years), if AGI can do almost all the work humans do now, traditional employment could shrink dramatically. Work itself would have to be redefined, and society would need big policy changes — like redistributing the wealth created by AI — just to stay stable. Most experts agree that AI will boost productivity and economic growth. Where they disagree is whether that growth will create more jobs than it destroys. The answer will depend on two big things:

1. How we steer the technology — whether we focus on augmenting humans or on pure replacement.
2. How smartly we build policies to manage the transition.

Next 5 years (2025–2030)

In the short term, AI — especially generative AI — is already starting to reshape jobs. It’s automating the routine parts of work and creating demand for new skills. If AGI-level systems emerge by 2030, the next five years won’t be about mass human replacement. Instead, we’ll see workplaces where humans and AI work side by side — augmented, not obsolete.

Most forecasts suggest we’ll see modest net changes in employment by 2030. There will be a lot of task disruption⁸— the way jobs get done will change — but overall job losses or gains will be limite⁹d. Here’s a table summarizing what’s most likely to happen across key dimensions during this first stage:

Medium term (5–20 Years: 2030–2045)

By 5 to 20 years out, we’ll be in the thick of AGI spreading through the world. If AGI shows up in the late 2020s or early 2030s, then between 2030 and 2045 we’ll see it fully take over the economy. This period will probably bring much bigger shifts in employment than anything we’ve seen so far.

The general expectation is that productivity will soar, and economic growth could speed up a lot as AI systems start to match or beat humans at most tasks¹⁰. But what happens to jobs is less clear. Most experts agree there will be a lot of job loss and a lot of new jobs created — they just don’t agree on which side will win out. Which way things go¹¹ depends on the path we take: do companies use AGI mostly to cut costs and replace workers, or do they use it to create new kinds of work and make people more powerful?

Here’s a sketch of what the medium-term might look like, assuming AGI becomes widely available and keeps getting better through the 2030s:

Long term (20+ Years, 2045 and beyond)

After 2045, things get even harder to predict. By then, AGI — and maybe even superintelligent AI — will be woven into every part of the economy. At that point, we have to assume AI can do almost any useful task¹² better than people¹³. That doesn’t mean humans stop working completely. But it does mean the old idea — that work is how you earn a living and how the economy runs — might be over.

What happens in this era will depend a lot on the choices we make earlier. In the best case, we spend the transition years building new systems, and by 2050 or 2060, people are living better than ever, with plenty of free time, thanks to AI-driven abundance. In the worst case, we ignore the risks, and AI-driven inequality tears society apart, leaving most people jobless and a few holding all the power.

The most likely future, based on what we know now, is somewhere in the middle: a world where human work shrinks and changes shape¹⁴, but doesn’t disappear, and where society has had to reinvent itself to keep up.

Across all these time horizons, there’s one thing everyone agrees on: AI — and especially AGI — will shake up labor markets around the world. In the next 5 years, we’ll see noticeable but manageable¹⁵ changes. Jobs will mostly be redefined¹⁶ rather than destroyed, with new opportunities popping up about as fast as automation cuts old ones.

In the medium term, once AGI really gets going, the disruption gets a lot bigger. Maybe a quarter or more of all jobs will be displaced or radically changed¹⁷. That will force societies to make big moves — building better safety nets and helping people retrain for new kinds¹⁸ of work.

By the long term, if AGI hits its full potential, human labor might no longer¹⁹ be the engine that drives the economy. Humans won’t become irrelevant²⁰, but society will have to organize itself around something other than paid work. Wealth will still be created — probably more than ever — but it will have to be distributed differently if we want a stable world.

Not everyone agrees on how this will play out. Some, like Aghion and his coauthors²¹, are optimistic. They think if we get the policies right, AI could boost productivity and create enough new jobs to keep employment strong. Others, like Acemoglu, Autor, and Susskind²², are more cautious. They warn that if we leave things to market forces, we could end up with fewer jobs and lower wages for most people — even if total economic output keeps growing²³.

The history of automation²⁴ backs up their concerns: it tends to make the rich richer unless something actively pushes back. Still, there’s wide agreement on one thing: outcomes aren’t fixed. They depend on what we do. Almost every expert points to the same fork in the road: will we use AI mostly to replace humans, or to augment them? Brynjolfsson calls this the choice between "automation" and "augmentation."²⁵

Regionally, rich countries²⁶ will probably feel the disruption first. Poorer countries might either leapfrog ahead by using AI, or get hurt if their traditional advantages disappear. Over the long run, there’s hope that AI’s enormous productivity could²⁷ lift living standards everywhere — the kind of future Sam Altman talks about²⁸, where AI growth funds a better life for everyone. But again, that will only happen if we make a real effort to share the gains.

If we don’t, we risk ending up with a world where a few people are on top and the rest are stuck with marginal jobs²⁹ or none at all. In short, work probably won’t vanish — but it will be transformed. In the next few years, the shifts will be big but manageable. Over the following decades, they’ll get deeper, forcing us to rethink how people earn a living and find meaning. By the late 21st century, paid work might not be the center of daily life anymore.

Whether that future is a golden age or a disaster depends on the choices we make now: whether we build AI to complement humans³⁰, invest in people, and update our social systems so the wealth created by machines lifts everyone up³¹.

Appendix: Staying useful

Limits to replacement

This is a a full summary, with discussion of limitations on the limits proposed by Daniel Susskind in his paper titled, “What Will Remain for People to Do?”

Hard to replace jobs

Footnotes

1

Wealthier countries generally report higher happiness, but here's the twist. happiness inequality often decreases even as income inequality rises. See the data.

2

Two leading economists warn AI could boost GDP while primarily replacing workers rather than enhancing their productivity. Acemoglu and Johnson call this the "just automate" trajectory—a path deepening inequality rather than shared prosperity.

3

Each technological revolution since the 1770s follows predictable phases. irruption, frenzy, crash, golden age, and maturity. Perez explains why new technologies take decades to fully impact productivity—organizations, skills, and supporting infrastructures must evolve first.

4

Humans don't lose jobs to technology—they shift to adjacent, complementary roles. Autor's work reveals this pattern across previous waves. The AI difference? It targets tasks once thought uniquely human.

5

The "Turing Trap" describes our economic risk when AI merely imitates humans rather than complementing them. When AI substitutes for labor, it weakens workers' bargaining power. Brynjolfsson argues that augmentation creates more total value while ensuring humans maintain economic leverage.

6

Even superintelligent machines would leave some economic niches for humans. Trade theory explains why—comparative advantage ensures specialization benefits both parties regardless of absolute skill differences.

7

Process-focused or outcome-focused? Early industrial revolutions created complementary opportunities for workers; recent digital waves primarily substitute for labor. The authors argue the current AI revolution could follow either path—it's a policy choice, not technological destiny.

8

Tracking AI's impact across 45 economies reveals a consistent pattern. Skill transformation trumps job elimination. Check the visuals.

9

$7 trillion GDP boost over a decade sounds impressive. Yet Goldman's models suggest this comes not from eliminating jobs but automating 25-50% of current work activities. Their analysis actually predicts significant productivity benefits for workers who effectively utilize the technology.

10

From "business as usual" to a genuinely transformative "new industrial revolution"—Korinek's most dramatic scenario envisions AI eventually handling all cognitive tasks. His research for the IMF suggests massive productivity gains but requires entirely new social contracts.

11

Seven plausible AI futures, from technologically stalled to widely deployed. Stanford's analysis emphasizes political choices over technological determinism—institutional decisions matter more than raw capabilities.

12

What work might humans keep even after widespread automation? Susskind identifies three enduring advantages. comparative economic advantages, preference for human services, and moral limitations. His framework anticipates areas where humans retain economic relevance despite expanding AI capabilities.

13

Imagine "the automation of virtually all tasks humans perform for pay." IMF researchers suggest this would fundamentally rewrite economic relationships.

14

Which jobs endure after AI becomes ubiquitous? Susskind explores three possibilities. those where human labor remains economically efficient, where people prefer human providers, or where ethics demand human oversight. The analysis provides a framework for predicting which roles might persist.

15

Specialized hybrid roles, not mass unemployment. WEF's visualizations track how technical skills increasingly augment domain knowledge rather than replacing it entirely.

16

Job automation versus task automation—a crucial distinction often overlooked. The research found 40% of work hours impacted but few jobs completely eliminated. AI typically changes how work happens rather than making entire roles obsolete.

17

As AI approaches human-level performance, labor displacement could accelerate dramatically—yet entirely new industries might emerge. IMF's modeling suggests outcomes depend heavily on policies promoting human-AI complementarity versus substitution.

18

78 million new jobs by 2030. The catch? 44% of workers' skills will be disrupted within five years. WEF warns this opportunity only materializes with massive upskilling efforts.

19

What happens when machines generate vast wealth with minimal human input? Traditional economic incentives could collapse without deliberate intervention. IMF scenarios explore this unsettling possibility.

20

Even if AI can handle almost everything, humans may reserve certain activities for themselves. Knight's analysis suggests this happens not because AI can't perform these tasks, but because we choose to keep them human for aesthetic or moral reasons.

21

AI will create more jobs than it destroys by unlocking entirely new industries. Aghion's team emphasizes that training, accessibility, and competition policies determine whether benefits are broadly shared or concentrated.

22

Current AI development favors replacement over enhancement. Leading economists argue we must redirect technology toward complementing workers rather than replacing them.

23

High-skill metro areas win; regions with less education lose. Autor's research shows technological benefits vary dramatically by geography, potentially widening regional inequality.

24

Industrial automation consistently increased productivity while wages stagnated for many. This pattern could intensify with AI unless balanced by deliberate redistribution mechanisms.

25

Imitation versus augmentation—a crucial distinction with profound consequences. The Stanford analysis shows how AI that substitutes for humans inherently weakens labor's bargaining position.

26

AI might allow wealthy countries to "re-shore" previously outsourced work through automation. MIT's Autor warns this could undermine traditional development paths for poorer nations, trapping some in middle-income status.

27

15-20% unemployment in advanced economies by late 2030s. Expert forecasts show significant uncertainty with some predicting much more severe disruption.

28

"Moore's Law for Everything." Altman envisions enormous productivity gains funding universal basic income through taxation on AI systems. His proposal suggests AI could dramatically lower prices while raising living standards—if benefits are properly distributed.

29

Current language models are being optimized to substitute for human labor rather than complement it. Acemoglu warns this path threatens millions of professional jobs while primarily benefiting technology owners.

30

Traditional safety nets may prove inadequate as automation accelerates. IMF research suggests entirely new social contracts might be needed.

31

Shared prosperity versus concentrated wealth—these divergent futures depend more on political choices than technological constraints. Historical patterns show similar technologies deployed with different institutional arrangements produce dramatically different outcomes

^{33233343536373839404142434445464748495051525354555657}