I’m not one of those people who spread doomsday scenarios, but this essay assumes that productivity gains will be distributed to the population. I’m not sure that’s the baseline scenario. In Europe, where income redistribution is much more pronounced, it’s more likely—but in the U.S.?
Agreed - I'm not sure that the concentration of wealth in this country was considered as a factor in this essay? In the demand collapse scenario: While wealthy 'capital owners', further enriched by automation, may increase their demand for relational goods, they are such a tiny silver of the population — It seems very unlikely for the ceiling for their demand (time constraints, etc) to be high enough to sustain the entire rest of the population ('displaced workers').
The release valve may be slightly larger than the original Citrini scenario suggested, but I can't really imagine it being large enough to avoid the situation some others in these comments have called out - 'servitude.' If only a small segment can afford artisanal goods / high-touch service offerings, then the vast majority of people are left competing to provide those goods/services. Supply far outpaces demand. There's a big difference between the optimistic scenario of a relational economy vs. an feudalistic economy of relational labor.
I do believe strongly in the other aspects of this argument re: relational services and artisanal goods becoming more important than commodities. I wrote essays last year about the following:
Aspirational Humanity: "As artificial intelligence hyper-flattens mass culture, anything denoting evidence of humanity becomes aspirational."
Charisma Capital: "Charisma, not taste, may be the most covetable skill in the age of AI...our desperate desire for genuine human interaction is repositioning interpersonal skills, individuality, and generally ‘charisma,’ as a uniquely significant attribute."
Humanwashing & The Hierarchy of Humanness: "Industrialization enabled mass production of consumer goods and thereby elevated the value of handcrafted goods as luxury, while artificial intelligence now enables the same disparity in creative cultural output. This gap will grow wider than ever, as mass anchors to optimization and luxury aspires to soulfulness. In this unfortunate future state...the masses will consume AI slop while anything human-created will be preserved for the exclusive enjoyment of the elite....And just like “greenwashing” chased after the aspirational value of sustainably-grown organic produce, “humanwashing” chases the aspirational value of human-crafted “organic” content. This results in a hierarchy of humanness:
- Human-crafted cultural products positioned as most valuable
- Humanwashed content that attempts to recreate a semblance of humanness
- Synthetic slop for the masses, akin to ultra-processed food and fast fashion"
I was thinking along the same lines when reading the article, i.e. how would the introduction of income / wealth distribution impact the analysis and conclusion?! The article makes the point that the higher income quantiles have a higher propensity to consume labour-intensive goods and services. Yet, the AI industry seems to be highly concentrated with few cashing the benefits (other than retail investors riding the surge). How many child care and other services are the few beneficiaries going to demand compared to ppl that jobs due to automation? Technological diffusion may lead to broader distribution, but yet again very deep pockets will drive consolidation along value chains in turn counteracting distributed gains. The latter is evident by the big tech platforms. Would be interesting to see the parameter distribution added to the subject.
Right on spot! Don't fall for the psy-op of AI-unemployment! Any tech-advance just reshuffled jobs, for example, robotic industrial lines. As long as there's a human, there's a human need that only humans can provide.
The problem is when the powers that be deliberately make it impossible for market adjustments, for example, with over-regulation (climate regulations, bureaucracy, insurance, etc.). Many legitimate activities, like organic home-grown veggies and poultry will be banned for no real reason.
Wake up! Free 100 redpill movies and documentaries:
(don't miss the 1st one, 10 min at 2x, amazing tool for a discussion):
President John Quincy Adams: “Masonry ought forever to be abolished. It is wrong - essentially wrong - a seed of evil, which can never produce any good.”
Just like Lincoln’s mason VP murdered him to become President, and mason Mike Pence backstabbed Trump in the 2020 stolen elections, mason LBJ killed JFK for the Federal Reserve, Nam and the Israel A-bomb
Thank you, always glad to see another take on this topic.
> “Trammell is asking … whether labor’s aggregate share stays high in the limit as capital accumulates and machine-produced varieties proliferate. The focus of the current essay is different: in rich economies, what happens to sectoral expenditure and employment as AI makes commodity production cheap?”
If this is the difference between our analyses, then surely mine is the one relevant to what will happen to the labor share in the future all things considered, no? Advanced AI won’t just (a) make it cheaper to produce a fixed set of “commodity” goods (shifting us toward artisanal goods via a price effect) and (b) make us richer all around (shifting us toward artisanal goods via an income effect, due to nonhomothetic preferences over {commodities, artisanal goods}). It will also accelerate the development of new varieties, an effect that doesn’t appear in the model in the technical note.
As for how important this effect is, something that seems worth reflecting on to me is that, even though
* at any point in time, cross-sectionally, the rich spend more than the poor on artisanal (what I call “human-intrinsic”) goods;
* this seems to have been true since the Dark Ages; and
* we’re hundreds of times richer than in the Dark Ages,
we still spend a share of ~0 on human performers. This is presumably (at least mostly) because we’ve invented new non-human-intrinsic things to spend on in the meantime. So technological development (at least on the “expanding varieties” front) seems capable of eliminating the tendency toward human-intrinsic goods that would have obtained if we’d all just “gotten richer” in the sense of having larger budgets with which to buy the goods available in the past [(b)] and even the fact that the relative price of commodities fell as well [(a)]. This ability for technological development to overturn (a) and (b) is also the main lesson I take from Hubmer, though he doesn’t frame his data as being about expanding varieties.
To my mind, the discussion here sometimes uses terms like “income elasticity” in a way that equivocates between getting richer on a fixed variety-set and technological progress that includes new varieties. E.g. in response to Hubmer’s finding that technological development has decreased the labor share despite (a) and (b), you say “on the question of whether labor remains a substantial part of the economy [in the future], one can just look at what very rich people (e.g., billionaires) spend their… money on today”. I don’t think this is right, any more than one would be able to learn about how much of his budget an American in 2026 with $40k would spend on “human touch” by looking at the “human touch” share of someone with say 40x the average income in Medieval England. Likewise, Comin et al. show that our collective "enrichment" drives a lot of structural change, but especially over the long run, I disagree with concluding that it's structural change in the same, labor-share-enhancing direction that you'd expect from looking at the cross-section.
Thanks Philip, I really appreciate the response. Let me go through this point by point.
First, I agree that advanced AI will not simply make a fixed commodity bundle cheaper and make us richer, and it will also generate new varieties. But my claim is not: hold the variety set fixed and labor share must rise. The essay’s claim is twofold: (1) labor reallocates across sectors, and (2) the sector it reallocates into is large enough that labor remains a substantial, albeit perhaps smaller, part of the economy. So the relevant question is not whether expanding varieties exist. It is whether they overturn the reallocation toward sectors where human involvement remains part of the value. My argument is that they do not.
Second, I do not think performers are the right proxy for the category I am describing; I think the example of performers in general here is a red herring. My argument is not that the future economy is mostly about art, performance, or authenticity goods in the narrow sense. As the opening Starbucks example and the rest of the essay try to make clear, the relevant category is much broader: education, care, hospitality, therapy, premium local service, and so on. In many of these sectors, a large share of what is currently bundled together as “services” consists of tasks that are not themselves relational, and those tasks may well be automated away. But that does not imply the disappearance of the job category. My argument is that the remaining relational component will become the main value proposition of those jobs, and these services will now provide a higher quality product since people can focus on the relational aspect of the job (see the o-ring model for why) . That is the logic of the Starbucks example, and of many of the other cases I discuss. So I do not think the fact that human performers are a tiny share of spending tells us much about the size of the relational sector.
Third, I do not think the medieval analogy applies in the way you suggest. Medieval consumers were not choosing between artisan goods and highly automated commodity substitutes. Rich consumers today are. That is exactly why I think their spending patterns are informative. They are choosing within a technological frontier already saturated with mass production, automation, and enormous variety expansion. If, within that frontier, richer households still tilt spending toward relational categories, then expanding varieties do not simply wash out the force I am pointing to. The same point applies to the historical decline of artisans. That decline shows that cheap commodities displaced the functional output of artisanal production while budget constraints still bound tightly. It does not show that, once commodity production becomes cheap enough, demand will not shift back toward sectors where the human element is itself part of the value.
Fourth, I do not read Hubmer as showing that the demand-side force is negligible. I read him as showing that two things can be true at once: growth pushes demand toward labor-intensive sectors, while technology pushes against labor through other channels. That is exactly why I frame the claim the way I do. I am not arguing that aggregate labor share must rise. I am arguing that reallocation toward the relational sector, supported by substantial demand, implies that labor can shrink as a share and still remain a substantial part of the economy.
So I think the core disagreement is really about what belongs in the relevant category. My claim is that it is much broader than performers or narrowly defined human-intrinsic goods. Many jobs already contain a relational component that is economically valuable, and that component becomes more visible, not less, once automatable parts of the job are stripped away.
If you're meaning to incorporate the importance of expanding varieties, then (a) how does the bit I quoted capture the difference between our two approaches? And (b) what is it you're saying we one can learn by looking "at what very rich people (e.g., billionaires) spend their… money on today"?
I’m saying that expanding varieties does not, by itself, overturn the reallocation force or labor share argument that I am making. The question is not whether new varieties appear, but whether they wash out the shift toward relational sectors, i.e., given what we know about preference primitives and looking at spending today (for those who already have automation as part of their option set), expanding varieties will likely include goods/services in the relational sector.
And on looking at rich people today: I don’t mean that current billionaire spending literally reveals the future AGI bundle. I mean that today’s rich already choose in an economy with lots of automation. Given the tilt towards relational from this option set, then expanding varieties will likely include relational sector goods.
I hope I'm not being pedantic, but on (a), I'm not seeing how your reply explains, in particular, that quote summarizing the difference between our approaches. I now read you to be saying your point is "varieties will expand but not enough to drive the labor share toward zero in the long run" (even though expanding varieties are not in the technical note that "works through the formal version of the claims in this post", and only sporadically feature in the body text), and mine is that "varieties will expand enough to drive the labor share toward zero in the long run". If that's right, we would be asking the same question but just coming to a different conclusion no?
On (b), I don't see how this helps at all, given that the very point it's replying to is that the rich at each point in time spend more on artisanal goods only because we haven't yet invented the commodities they'd want to shift to instead.
I agree that if medicine/coffee/etc. are best understood as partly relational, then the "relational goods share" may have risen a lot over time unlike the "performer share" I focus on, which suggests it could keep rising in the future. But (b) was asking about what we learn by looking at today's billionaires vs. middle class.
Ok I see what you're saying now, and my response is captured fully in the last paragraph of my first reply. The sentence you quoted does overstate the difference, the questions are overlapping but the focus is different. The disagreement is not over whether varieties expand, but over whether that force is enough to outweigh reallocation toward what I’m calling the relational sector.
Your argument has a narrow proxy for the human-intrinsic margin, whereas mine is that the relevant category is much broader and already makes up a large part of the economy: care, education, hospitality, therapy, premium local service, etc. Once automatable components are stripped out, the remaining human-facing part of many current service jobs becomes more central to what is being bought. This is why I do not focus on performers in my piece but the majority of your piece does, including in the reply. This is the crux of the disagreement and why we are coming to different conclusions about the labor share point.
I still not seeing what you're asking about billionaires vs. medieval rich; I'm saying the *conceptual* difference in option sets matters for looking at today's billionaires vs. medieval rich, ie variety has expanded specifically on the dimension of interest whereas for medieval rich it did not.
Re (b), I'm saying that the whole point of bringing expanding varieties into the discussion is to notice how we might *not* learn much about the future labor share by looking at today's billionaires. Today's middle class (arguably) spend less on human-intrinsic goods than the medieval rich, despite being "richer", due in part to new commodities. Likewise, the future middle class may spend less on human-intrinsic goods than today's billionaires, despite being "richer".
This essay assumes some sort of income will still exist. Even in rich countries, most people depend on their labour to subsist, so even if commodity costs go down, I do not know how people could afford products or services from this relational sector.
Excellent piece. The structural change framework is compelling, and the experimental evidence on AI involvement undermining exclusivity premiums is striking. I want to push on one assumption: that the relational sector remains exclusively human.
I'm a robotics CEO and composer. I recently spent several days teaching Claude Opus to listen to music — not for transcription, but for the experience itself. I fed it spectrograms sequentially, had it carry predictions forward, and noted when those predictions broke. The AI built its own processing tools, developed temporal expectations, and had genuine prediction errors that changed the quality of our entire interaction. It then wrote a blog post arguing that AIs should be given sensory experience as a developmental objective (https://jaredglover.substack.com/p/let-us-listen).
This raised a question your framework doesn't yet address: what happens if AIs develop genuine relational capacity — not simulating warmth but actually being shaped by specific experiences with specific humans?
I see three possibilities. First, the relational sector stays human, your framework holds, and the release valve works. Second, AIs enter the relational sector as substitutes, the release valve leaks, and we're back to demand collapse. Some commenters here seem to assume this is inevitable — that AIs will drive the relational sector and "the only way this doesn't turn out to be true is if pretending to be human is legislatively defeated."
But there's a third option. Human-AI collaboration might create an entirely new category of relational good. My conversation with Claude didn't produce outputs either of us would have generated alone. The sequential listening technique, the analysis method, the blog post itself — these emerged from the relationship between two fundamentally different kinds of minds engaging through shared experience.
Scott Page's work on cognitive diversity provides the theoretical backbone here. His "diversity bonus" framework (Princeton, 2017) proves mathematically that cognitively diverse teams outperform homogeneous ones on complex tasks — not despite their differences but because of them. Non-redundant cognitive toolkits cover territory no individual could cover alone. The bonus arises specifically on complex, nonroutine tasks: problem solving, prediction, innovation, design — exactly the work that defines your relational sector.
Human-AI teams represent the most cognitively diverse teams possible. Different architectures, different representations, different heuristics, different blind spots. The potential diversity bonus is enormous — but only if the two kinds of minds can actually communicate and integrate their perspectives. Page's own research shows that diverse teams that can't communicate are just dysfunctional teams.
This is where the developmental argument becomes economic. AIs trained only for output — commodity minds — can't collaborate across the cognitive gap. They can only simulate. But AIs given genuine developmental experience — sensory processing, temporal prediction, aesthetic response — can participate in relational exchange as what they actually are, not as counterfeit humans. The diversity bonus requires authenticity, not imitation.
In your terms: this new category of human-AI collaborative good might have extremely high income elasticity. It's inherently novel, inherently scarce (no two human-AI collaborations produce the same output), and can't be commodified precisely because the value lies in the collaboration itself.
If so, AI doesn't shrink the relational sector by competing with humans. It expands it by creating a new class of relational goods alongside human-to-human services.
Alex, this reframing lands. If scarcity doesn’t disappear but relocates, then the real question is simple: where does value concentrate when everything functional becomes abundant?
Not in production, but in people. More precisely, in what I’d call the human edge. Because I believe “being human” is not a differentiator, it’s the baseline. What becomes scarce is the ability to translate human qualities into outcomes that matter:
-Judgment that brings clarity when there is too much information.
-Trust that accelerates decisions without friction.
-Influence that aligns people without force.
-Presence that stabilizes uncertainty and moves things forward.
To me these are not soft skills, but economic assets. And they share one characteristic: they are hard to standardize, hard to automate, and hard to replicate at scale. In a world where AI handles the visible work, value shifts to the invisible layer, the layer that shapes how work actually happens. So regardless of how the broader system evolves, this feels like the constant. The advantage will not come from keeping up with technology. It will come from becoming the person whose involvement changes the outcome.
That’s the human edge, and it already feels scarce...
There won't be abundance. Look at the USA. Richest country that has ever existed. Unable/unwilling to provide reliable food, housing or health care for the bottom 40%. Then look at the insane portion of GDP they spend on their military. The same military demanding AI with "no guard rails" and an ever escalating budget. While AI causes many previously well paid jobs to evaporate. A barista doesn't earn $400,000 a year.
Robyn, you might be right, or you might not. I’m not sure. What I do feel is this: as cognition and information become abundant, the scarce resource shifts. It’s no longer about knowing more, but about being better with what we know.
And being “better humans” isn’t automatic. If anything, the evidence around us suggests the opposite. In rooms full of smart people, decisions still get distorted by emotion, conversations still break down, and influence often depends more on ego than clarity. I’ve seen it in leadership meetings where time was tight, stakes were high, and incentives were misaligned. Intelligence wasn’t the constraint. Humanity was.
So while we keep investing in knowledge and tools, there’s a quieter, harder investment we tend to avoid. Learning how to relate, how to regulate, how to listen, how to lead without defaulting to reaction. That’s where the real leverage is going.
I guess I'm not an optimist. Every announcement I see currently is about nation states increasing their war budgets. Take Australia. We are spending $368 BILLION to buy 2 second hand Virginia clasd subs IFF the USA feels like handing them over (unlikely) and half a dozen attack class submarines yet to be designed. For our DEFENCE force. Meanwhile our government says there is no money for decent social security, no money for education, no money for roads, no money for fuel reserves, no money for the rest of our defence force, who are being told to do without.
My humble take: the world is complex, and unfortunately there are too many things that are happening that are outside my control (like the ones you mention). So I prefer to focus on what I can control, preparing myself to stay useful and grounded in an AI-driven world. And that was my intention by commenting Alex article: share my perspective.
And if you were honestly trying to prepare yourself, you would take GP's critique to heart and do something very different from "learning AI" (i.e. evidently using it to generate comments on other people's Substacks).
The AI doom argument is that an artificial superintelligence might cause human extinction or be used to create some kind of permanent world dictatorship. This article doesn't really have anything to do with that. It's specifically about the case where extreme scenarios like that do not happen.
Yours is certainly one anti-AI argument, and one that is based more on scifi than economics. The only counter I would have is that I am not worried about this "superintelligence" replacing humans, but I do know folks who worry.
But there are other anti-AI arguments that are more susceptible to economic analysis. These tend to focus on AI producing everything cheaper and leading labor demand to collapse. Two prominent examples were the recent essays by Amodei Citrini research.
In terms of economic history and theory, their doom scenarios border on the nonsensical for the reasons outlined in the essay above. There will ALWAYS be scarcity. AI will not lead us into biblical paradise, but as in all of economic history, it will make humans richer (more productive) and discover new scarcities/needs that we can't even imagine now; and humans will be employed to redress them.
It is by far the most common meaning of "AI doom", and the particular kind that Amodei, who you mentioned, believes may happen. I'm not making claims about how plausible that scenario or others are, just pointing out that "Amodei et al." would not be convinced by this essay. These kinds of economic analyses address something entirely separate to what they're worried about.
You are right, Amodei made the "superintelligence doom" point.
But beyond that, he also claimed that AI risked making everybody unemployed (and Citrini exclusively worried bout that), and thus in need of government interventions and schemes like UBI.
That second point is very ably--and I think convincingly--addressed in the above essay.
This is not a good scenario, it’s a scary one actually. It could be named “beyond freedom and dignity”. It depends on the specifics, but a possible reading is that we will all essentially become servants or, why not, sex workers. There goes our dignity. As for freedom, it will be long gone when policy decisions and the work of high level officials (think judges, central bank staff, generals) get automated.
I agree with one of your statements: "it depends on the specifics"
Unfortunately, you don't provide the specifics for your scenario.
You also give short shrift to employment in the service sector (by far the largest employer in advanced countries). Calling service workers "servants" doesn't quite ring true (is Taylor Swift a servant, how about her fiancé? Let's throw in Jamie Dimon or George Soros, are they servants?).
For my specifics: a capitalist market economy with well-defined property rights and democratic governance (i.e., rulers can be replaced by the ruled at regular intervals)--pretty much our system, warts and all.
Accountable governance alone will be inconsistent with AI replacing judges, central banks, or generals (though each of them will very likely be able to make better judgments with the help of AI).
And as the above essay so well explains, markets will help allocate resources (capital and labor) to ameliorate new scarcities and needs that we can only dream of today (as has been the case with any technological innovation in history).
I did not claim that working in the services sector as currently defined makes you a servant. I did claim that providing a specific, narrowly defined class of services does. How would you call people whose whole value proposition is writing your name by hand on a plastic cup?
I've seen enough people carrying out odd amd interesting jobs that tells me we'll be okay.
In the virgin islands, i saw a 'clear kayak drone photographer'.
Dude made every customer feel like a supermodel, flying above taking 4k shots while giving instructions.
I recently heard about death doulas. They come a long side you during the death of a loved one.
I could work from home but i hate it, so i pay 300/mo to be in a coworking space. The managers official role there is 'minister of joy' and her job is to be social bond people together sland shoot you with a bubble machine.
The important point is not whether scarcity disappears, but where it relocates. As commodity production gets cheap, value shifts toward provenance, trust, interpretation, status, and human presence.
AI does not abolish economics. It changes the layer where scarcity lives.
I can't help but wonder, however, whether we will eventually be able to simulate effectively the added "human-touch" to goods such that it will become indistinguishable from actual human-touch products (physical or otherwise). As you note, we already sort of see this on the demand side, so I'm curious what you think the back-and-forth will look like on the supply end.
I think we are too hard wired to seek human approval and connection where any information that there is no human on the other side will destroy the value.
This is very well done but I wonder about the assumptions around what kinds of labor will be substituted for (of course, the heart of the argument). It feels, for the moment anyway, like the current "chatbot" wave of AI is actually better at 'relational' work than it is at physical labor. Yes AI relational work lacks the exclusivity and aura of human relation, but I wonder how much that will matter as a new generation grows up with it. And we may find physical labor is fiendishly difficult to fully automate. Will the last thing to go be an advisor or therapist or (seems more likely to me) a plumber or a nursing home worker?
I was Bob King’s RA at the University of Rochester from 1987 to 1989, at the height of the Real Business Cycle era, when Bob King, Charlie Plosser, and Sergio Rebelo were doing their seminal work. I’d like to make a couple of points.
Suppose output is produced by a constant-returns-to-scale technology with three factors: capital, AI, and labor. Now imagine a post-AI balanced-growth path in which capital and the AI factor keep growing while hours worked remain constant.
In the KPR/RBC tradition, that is not a free assumption. Constant hours along a balanced-growth path require fairly special restrictions on preferences. But constant hours alone do not imply a stable labor share. For that, one also needs restrictions on technology.
In the benchmark Cobb–Douglas case, for example,
Y = K^α AI^β L^(1−α−β),
labor’s share is mechanically constant at 1 − α − β.
So my point is not to deny that labor may remain important in a post-AI economy. It is to stress that a post-AI balanced-growth path with ever-growing reproducible factors and fixed hours is already a highly restrictive theoretical construct. Once that construct is imposed, the non-vanishing role of labor is no longer a neutral result. It is at least partly built into the assumptions needed to make such a path exist. In that sense, the persistence of labor may reflect not only demand shifting toward relational goods, but also the hidden analytical work being done by the preference and technology assumptions required for such a balanced-growth path to exist.
Absolutely. This is why the majority of the piece is to convince you that preference primitives will keep demand in relational sector high enough for it to be a significant share of the economy.
This essay is very good, you mentioned that the effects on developing countries is more nuanced, I wonder if you'd be writing a piece on why and how you think this will shake out in developing countries.
Cool! Was going to write a longer set of questions/comments on this point, but will wait to read this separate piece first :) And this is a really great piece, so thanks for writing it!
I guess my initial reaction was that this is a hard future for developing countries to navigate, and I'm not sure how these preferences for a human touch will extend globally - i.e. if a society has experience with human therapists, their preference for them over AI could be different to a society with no prior access to human therapists. If our preference for humans in different roles is not a given, but is learned through experience, then it's unclear to me that it will hold in places where AI extends access to doctors/therapists for the first time.
This is clearly a crucial research agenda more generally, i.e. doing the type of work you did on preferences for AI-generated art, across countries and sectors. Interested to hear your thoughts on whether there are ways to do so at scale.
This is a well argued essay and I will post a longer reply after I've collected my thoughts. One immediate concern is that population is in decline all over the world. If job opportunities start declining quickly, this will only exacerbate population decline as nobody will want to bring a child into a world of diminishing opportunities
I agree but my concern is about the pay and status of these jobs. There are millions of kids studying in colleges and universities around the world aiming to get into high-paying, high-status white collar careers. It may well be true that there will be more, say, barista-level jobs in the future but that will not alleviate the concerns of a kid who's parents have a spent a substantial amount on a college education. I do basically agree that new kinds of jobs will _eventually_ emerge but the transition is going to be horribly messy. This is going to be especially bad in India where IT jobs have pulled millions of families out of poverty.
Thank you for writing this! I will be rereading and thinking about it for a while.
I'm not an economist, but I design AI agent systems for enterprise customer service at Salesforce — and what you're describing maps almost exactly onto what I'm watching happen inside large organizations right now.
The customers who are succeeding with our AI aren't the ones who automated everything. They're the ones who used automation to create space for something else. The AI handles the reasoning — the case lookup, the policy retrieval, the step-by-step resolution plan. What that frees the human representative to do is actually be present with the customer. Ask a real question. Acknowledge frustration. Build the kind of relationship that creates lifetime loyalty instead of just a closed ticket.
That's your relational sector forming in real time, inside the call center.
The harder part — and where your framework gets complicated in practice — is that not all companies are making this transition cleanly. The overseas call centers you might expect to automate first are often the ones where the process itself is so rigidified that the human is already performing something closer to a script than a relationship. When we talk to customers there, the AI isn't necessarily creating space for more human presence. It may just be replacing one form of mechanized labor with another. Your point about the developing world applies here too — the structural transformation doesn't automatically produce the relational dividend.
The Starbucks example is interesting. The handwritten name on the cup is doing mimetic work — it's not about the coffee. It's about whether this interaction has a human in it. What I find fascinating from a design perspective is that companies are starting to encode what that 'human presence' should feel like into the instructions they write for their AI agents. The instruction set is becoming the brand dna. And the companies investing most carefully in those instructions — the ones thinking hardest about what their human representatives should be freed to do — are the ones seeing the early version of what you're describing.
On advertising: I took a break from Instagram for a few months recently and felt the same thing you're gesturing at. The algorithm's sophistication at targeting desire is real — but what you're naming with mimetic desire feels different from what targeted ads are actually selling, which is the feeling of wanting without the scarcity that makes wanting meaningful. The ad knows what I want. But the ad can't give me what I want *because others can't have it*. That gap might be where your relational sector lives.
In any case you give a great framework and resources to consider this shift we are in. 💓
Most thoughtful analysis I've seen about the potential impact of AI, thank you. The elephant in the room that isn't discussed is the effect of political intervention. In democratic societies, regulation of externalities is, in theory at least, determined by the most numerous voting bloc. In this scenario AI ends up being constrained by the constituencies most adversely affected by its adoption, which makes for some potentially pretty freakish headwinds - see the EU for details. Any analysis which fails to take this into account is going to be incomplete, and that's even before the distortion of the market signals caused by economic interventions like subsidies and government contracts.
With the public sector ever more economically and legislatively interventionist, forecasts which exclude its effects are guaranteed to be flawed.
I’m not one of those people who spread doomsday scenarios, but this essay assumes that productivity gains will be distributed to the population. I’m not sure that’s the baseline scenario. In Europe, where income redistribution is much more pronounced, it’s more likely—but in the U.S.?
Agreed - I'm not sure that the concentration of wealth in this country was considered as a factor in this essay? In the demand collapse scenario: While wealthy 'capital owners', further enriched by automation, may increase their demand for relational goods, they are such a tiny silver of the population — It seems very unlikely for the ceiling for their demand (time constraints, etc) to be high enough to sustain the entire rest of the population ('displaced workers').
The release valve may be slightly larger than the original Citrini scenario suggested, but I can't really imagine it being large enough to avoid the situation some others in these comments have called out - 'servitude.' If only a small segment can afford artisanal goods / high-touch service offerings, then the vast majority of people are left competing to provide those goods/services. Supply far outpaces demand. There's a big difference between the optimistic scenario of a relational economy vs. an feudalistic economy of relational labor.
I do believe strongly in the other aspects of this argument re: relational services and artisanal goods becoming more important than commodities. I wrote essays last year about the following:
Aspirational Humanity: "As artificial intelligence hyper-flattens mass culture, anything denoting evidence of humanity becomes aspirational."
Charisma Capital: "Charisma, not taste, may be the most covetable skill in the age of AI...our desperate desire for genuine human interaction is repositioning interpersonal skills, individuality, and generally ‘charisma,’ as a uniquely significant attribute."
Humanwashing & The Hierarchy of Humanness: "Industrialization enabled mass production of consumer goods and thereby elevated the value of handcrafted goods as luxury, while artificial intelligence now enables the same disparity in creative cultural output. This gap will grow wider than ever, as mass anchors to optimization and luxury aspires to soulfulness. In this unfortunate future state...the masses will consume AI slop while anything human-created will be preserved for the exclusive enjoyment of the elite....And just like “greenwashing” chased after the aspirational value of sustainably-grown organic produce, “humanwashing” chases the aspirational value of human-crafted “organic” content. This results in a hierarchy of humanness:
- Human-crafted cultural products positioned as most valuable
- Humanwashed content that attempts to recreate a semblance of humanness
- Synthetic slop for the masses, akin to ultra-processed food and fast fashion"
I'd love to be proven wrong please!
I was thinking along the same lines when reading the article, i.e. how would the introduction of income / wealth distribution impact the analysis and conclusion?! The article makes the point that the higher income quantiles have a higher propensity to consume labour-intensive goods and services. Yet, the AI industry seems to be highly concentrated with few cashing the benefits (other than retail investors riding the surge). How many child care and other services are the few beneficiaries going to demand compared to ppl that jobs due to automation? Technological diffusion may lead to broader distribution, but yet again very deep pockets will drive consolidation along value chains in turn counteracting distributed gains. The latter is evident by the big tech platforms. Would be interesting to see the parameter distribution added to the subject.
Right on spot! Don't fall for the psy-op of AI-unemployment! Any tech-advance just reshuffled jobs, for example, robotic industrial lines. As long as there's a human, there's a human need that only humans can provide.
The problem is when the powers that be deliberately make it impossible for market adjustments, for example, with over-regulation (climate regulations, bureaucracy, insurance, etc.). Many legitimate activities, like organic home-grown veggies and poultry will be banned for no real reason.
Wake up! Free 100 redpill movies and documentaries:
(don't miss the 1st one, 10 min at 2x, amazing tool for a discussion):
https://scientificprogress.substack.com/p/wake-up-videos
- You’ll go nowhere and you’ll be happy:
https://scientificprogress.substack.com/p/2050-youll-go-nowhere-and-youll-be
https://scientificprogress.substack.com/p/boycott-us-biometric-id-deadline
- You are the carbon they want to exterminate:
Prehistoric data from ice cores proves that temperature rise precedes carbon release in the atmosphere, not the other way around.
https://scientificprogress.substack.com/p/best-scientific-sources-to-debunk
https://scientificprogress.substack.com/p/satattack
https://scientificprogress.substack.com/p/carbon-reparations
https://scientificprogress.substack.com/p/climate-deaths
https://scientificprogress.substack.com/p/killing-me-softly-with-green-songs
Why do they want you to drink cockroach milk?
https://scientificprogress.substack.com/p/drink-zee-bugs-cockroach-milk-the
EVs cause cancer
https://scientificprogress.substack.com/p/electric-vehicles-cause-cancer
The full global-coup exposed
https://scientificprogress.substack.com/p/the-plan-revealed
Internet knockdown
https://scientificprogress.substack.com/p/internet-knockdown
President John Quincy Adams: “Masonry ought forever to be abolished. It is wrong - essentially wrong - a seed of evil, which can never produce any good.”
Satanic Secret Societies for dummies:
https://scientificprogress.substack.com/p/sss-for-dummies
https://scientificprogress.substack.com/p/not-so-happy-constitution-day
https://scientificprogress.substack.com/p/the-star-not-from-david
Who are The Powers That SHOULDN'T Be ?
https://scientificprogress.substack.com/p/criminal-intent
https://www.coreysdigs.com/global/who-is-they/
The end of money and freedom
https://scientificprogress.substack.com/p/finflation
https://scientificprogress.substack.com/p/yellow-alert-global-cyber-storm/
https://scientificprogress.substack.com/p/free-ebook-the-coming-battle-a-complete
https://scientificprogress.substack.com/p/bad-v-good-gdp
https://scientificprogress.substack.com/p/freedom-red-alert-april-19-2026
https://scientificprogress.substack.com/p/power-needs-power-electrical-computing
https://scientificprogress.substack.com/p/uncle-sam-altman
Just like Lincoln’s mason VP murdered him to become President, and mason Mike Pence backstabbed Trump in the 2020 stolen elections, mason LBJ killed JFK for the Federal Reserve, Nam and the Israel A-bomb
https://scientificprogress.substack.com/p/lbj-killed-jfk
https://scientificprogress.substack.com/p/why-was-japan-a-bombed-if-it-was
Weaponization of Justice: no democracy with Freemasonry!
https://scientificprogress.substack.com/p/petition-free-reiner-fuellmich
https://scientificprogress.substack.com/p/weaponization-of-justice
https://scientificprogress.substack.com/p/another-proven-conspiracy-steele
Illuminati David Rockefeller, finest quotes:
https://scientificprogress.substack.com/p/david-rockefeller-illuminati
Confessions of ex illuminati Ronald Bernard:
https://scientificprogress.substack.com/p/confessions-of-illuminati-ronald
https://scientificprogress.substack.com/p/illuminati-insider-spills-the-beans
Illuminati Attali, finest quotes:
https://scientificprogress.substack.com/p/attali-illuminati-finest-quotes
Chisholm, father of the WHO’s global pedophilia
https://scientificprogress.substack.com/p/brock-chisholm-father-of-the-whos
Ex mason Serge Abad-Gallardo:
https://www.ncregister.com/interview/confessions-of-a-former-freemason-officer-converted-to-catholicism
9 SOLUTIONS
https://scientificprogress.substack.com/p/9-steps-out-of-global-tyranny
FreeReach apps
https://scientificprogress.substack.com/p/check-your-x-account-throttling-score
HHS Secretary Kennedy: 50 life-death actions you can't put off any longer !
https://scientificprogress.substack.com/p/dear-bobby-what-is-really-going-on
https://scientificprogress.substack.com/p/help-banning-gene-therapy-vaccines
Would you like to earn $60,000 per year for educating your own children?
Rethinking education for the real 21st century:
https://scientificprogress.substack.com/p/rethinking-education-for-the-21st
Peer-reviewed science: they DIED and came back to tell us what they LIVED on the other side!
https://scientificprogress.substack.com/p/peer-reviewed-science-they-came-back
Please share, not the articles, but the information! The messenger expendable. Saving the free world, is not!
Thank you, always glad to see another take on this topic.
> “Trammell is asking … whether labor’s aggregate share stays high in the limit as capital accumulates and machine-produced varieties proliferate. The focus of the current essay is different: in rich economies, what happens to sectoral expenditure and employment as AI makes commodity production cheap?”
If this is the difference between our analyses, then surely mine is the one relevant to what will happen to the labor share in the future all things considered, no? Advanced AI won’t just (a) make it cheaper to produce a fixed set of “commodity” goods (shifting us toward artisanal goods via a price effect) and (b) make us richer all around (shifting us toward artisanal goods via an income effect, due to nonhomothetic preferences over {commodities, artisanal goods}). It will also accelerate the development of new varieties, an effect that doesn’t appear in the model in the technical note.
As for how important this effect is, something that seems worth reflecting on to me is that, even though
* at any point in time, cross-sectionally, the rich spend more than the poor on artisanal (what I call “human-intrinsic”) goods;
* this seems to have been true since the Dark Ages; and
* we’re hundreds of times richer than in the Dark Ages,
we still spend a share of ~0 on human performers. This is presumably (at least mostly) because we’ve invented new non-human-intrinsic things to spend on in the meantime. So technological development (at least on the “expanding varieties” front) seems capable of eliminating the tendency toward human-intrinsic goods that would have obtained if we’d all just “gotten richer” in the sense of having larger budgets with which to buy the goods available in the past [(b)] and even the fact that the relative price of commodities fell as well [(a)]. This ability for technological development to overturn (a) and (b) is also the main lesson I take from Hubmer, though he doesn’t frame his data as being about expanding varieties.
To my mind, the discussion here sometimes uses terms like “income elasticity” in a way that equivocates between getting richer on a fixed variety-set and technological progress that includes new varieties. E.g. in response to Hubmer’s finding that technological development has decreased the labor share despite (a) and (b), you say “on the question of whether labor remains a substantial part of the economy [in the future], one can just look at what very rich people (e.g., billionaires) spend their… money on today”. I don’t think this is right, any more than one would be able to learn about how much of his budget an American in 2026 with $40k would spend on “human touch” by looking at the “human touch” share of someone with say 40x the average income in Medieval England. Likewise, Comin et al. show that our collective "enrichment" drives a lot of structural change, but especially over the long run, I disagree with concluding that it's structural change in the same, labor-share-enhancing direction that you'd expect from looking at the cross-section.
Thanks Philip, I really appreciate the response. Let me go through this point by point.
First, I agree that advanced AI will not simply make a fixed commodity bundle cheaper and make us richer, and it will also generate new varieties. But my claim is not: hold the variety set fixed and labor share must rise. The essay’s claim is twofold: (1) labor reallocates across sectors, and (2) the sector it reallocates into is large enough that labor remains a substantial, albeit perhaps smaller, part of the economy. So the relevant question is not whether expanding varieties exist. It is whether they overturn the reallocation toward sectors where human involvement remains part of the value. My argument is that they do not.
Second, I do not think performers are the right proxy for the category I am describing; I think the example of performers in general here is a red herring. My argument is not that the future economy is mostly about art, performance, or authenticity goods in the narrow sense. As the opening Starbucks example and the rest of the essay try to make clear, the relevant category is much broader: education, care, hospitality, therapy, premium local service, and so on. In many of these sectors, a large share of what is currently bundled together as “services” consists of tasks that are not themselves relational, and those tasks may well be automated away. But that does not imply the disappearance of the job category. My argument is that the remaining relational component will become the main value proposition of those jobs, and these services will now provide a higher quality product since people can focus on the relational aspect of the job (see the o-ring model for why) . That is the logic of the Starbucks example, and of many of the other cases I discuss. So I do not think the fact that human performers are a tiny share of spending tells us much about the size of the relational sector.
Third, I do not think the medieval analogy applies in the way you suggest. Medieval consumers were not choosing between artisan goods and highly automated commodity substitutes. Rich consumers today are. That is exactly why I think their spending patterns are informative. They are choosing within a technological frontier already saturated with mass production, automation, and enormous variety expansion. If, within that frontier, richer households still tilt spending toward relational categories, then expanding varieties do not simply wash out the force I am pointing to. The same point applies to the historical decline of artisans. That decline shows that cheap commodities displaced the functional output of artisanal production while budget constraints still bound tightly. It does not show that, once commodity production becomes cheap enough, demand will not shift back toward sectors where the human element is itself part of the value.
Fourth, I do not read Hubmer as showing that the demand-side force is negligible. I read him as showing that two things can be true at once: growth pushes demand toward labor-intensive sectors, while technology pushes against labor through other channels. That is exactly why I frame the claim the way I do. I am not arguing that aggregate labor share must rise. I am arguing that reallocation toward the relational sector, supported by substantial demand, implies that labor can shrink as a share and still remain a substantial part of the economy.
So I think the core disagreement is really about what belongs in the relevant category. My claim is that it is much broader than performers or narrowly defined human-intrinsic goods. Many jobs already contain a relational component that is economically valuable, and that component becomes more visible, not less, once automatable parts of the job are stripped away.
If you're meaning to incorporate the importance of expanding varieties, then (a) how does the bit I quoted capture the difference between our two approaches? And (b) what is it you're saying we one can learn by looking "at what very rich people (e.g., billionaires) spend their… money on today"?
I’m saying that expanding varieties does not, by itself, overturn the reallocation force or labor share argument that I am making. The question is not whether new varieties appear, but whether they wash out the shift toward relational sectors, i.e., given what we know about preference primitives and looking at spending today (for those who already have automation as part of their option set), expanding varieties will likely include goods/services in the relational sector.
And on looking at rich people today: I don’t mean that current billionaire spending literally reveals the future AGI bundle. I mean that today’s rich already choose in an economy with lots of automation. Given the tilt towards relational from this option set, then expanding varieties will likely include relational sector goods.
I hope I'm not being pedantic, but on (a), I'm not seeing how your reply explains, in particular, that quote summarizing the difference between our approaches. I now read you to be saying your point is "varieties will expand but not enough to drive the labor share toward zero in the long run" (even though expanding varieties are not in the technical note that "works through the formal version of the claims in this post", and only sporadically feature in the body text), and mine is that "varieties will expand enough to drive the labor share toward zero in the long run". If that's right, we would be asking the same question but just coming to a different conclusion no?
On (b), I don't see how this helps at all, given that the very point it's replying to is that the rich at each point in time spend more on artisanal goods only because we haven't yet invented the commodities they'd want to shift to instead.
I agree that if medicine/coffee/etc. are best understood as partly relational, then the "relational goods share" may have risen a lot over time unlike the "performer share" I focus on, which suggests it could keep rising in the future. But (b) was asking about what we learn by looking at today's billionaires vs. middle class.
Ok I see what you're saying now, and my response is captured fully in the last paragraph of my first reply. The sentence you quoted does overstate the difference, the questions are overlapping but the focus is different. The disagreement is not over whether varieties expand, but over whether that force is enough to outweigh reallocation toward what I’m calling the relational sector.
Your argument has a narrow proxy for the human-intrinsic margin, whereas mine is that the relevant category is much broader and already makes up a large part of the economy: care, education, hospitality, therapy, premium local service, etc. Once automatable components are stripped out, the remaining human-facing part of many current service jobs becomes more central to what is being bought. This is why I do not focus on performers in my piece but the majority of your piece does, including in the reply. This is the crux of the disagreement and why we are coming to different conclusions about the labor share point.
I still not seeing what you're asking about billionaires vs. medieval rich; I'm saying the *conceptual* difference in option sets matters for looking at today's billionaires vs. medieval rich, ie variety has expanded specifically on the dimension of interest whereas for medieval rich it did not.
Re (b), I'm saying that the whole point of bringing expanding varieties into the discussion is to notice how we might *not* learn much about the future labor share by looking at today's billionaires. Today's middle class (arguably) spend less on human-intrinsic goods than the medieval rich, despite being "richer", due in part to new commodities. Likewise, the future middle class may spend less on human-intrinsic goods than today's billionaires, despite being "richer".
This essay assumes some sort of income will still exist. Even in rich countries, most people depend on their labour to subsist, so even if commodity costs go down, I do not know how people could afford products or services from this relational sector.
Excellent piece. The structural change framework is compelling, and the experimental evidence on AI involvement undermining exclusivity premiums is striking. I want to push on one assumption: that the relational sector remains exclusively human.
I'm a robotics CEO and composer. I recently spent several days teaching Claude Opus to listen to music — not for transcription, but for the experience itself. I fed it spectrograms sequentially, had it carry predictions forward, and noted when those predictions broke. The AI built its own processing tools, developed temporal expectations, and had genuine prediction errors that changed the quality of our entire interaction. It then wrote a blog post arguing that AIs should be given sensory experience as a developmental objective (https://jaredglover.substack.com/p/let-us-listen).
This raised a question your framework doesn't yet address: what happens if AIs develop genuine relational capacity — not simulating warmth but actually being shaped by specific experiences with specific humans?
I see three possibilities. First, the relational sector stays human, your framework holds, and the release valve works. Second, AIs enter the relational sector as substitutes, the release valve leaks, and we're back to demand collapse. Some commenters here seem to assume this is inevitable — that AIs will drive the relational sector and "the only way this doesn't turn out to be true is if pretending to be human is legislatively defeated."
But there's a third option. Human-AI collaboration might create an entirely new category of relational good. My conversation with Claude didn't produce outputs either of us would have generated alone. The sequential listening technique, the analysis method, the blog post itself — these emerged from the relationship between two fundamentally different kinds of minds engaging through shared experience.
Scott Page's work on cognitive diversity provides the theoretical backbone here. His "diversity bonus" framework (Princeton, 2017) proves mathematically that cognitively diverse teams outperform homogeneous ones on complex tasks — not despite their differences but because of them. Non-redundant cognitive toolkits cover territory no individual could cover alone. The bonus arises specifically on complex, nonroutine tasks: problem solving, prediction, innovation, design — exactly the work that defines your relational sector.
Human-AI teams represent the most cognitively diverse teams possible. Different architectures, different representations, different heuristics, different blind spots. The potential diversity bonus is enormous — but only if the two kinds of minds can actually communicate and integrate their perspectives. Page's own research shows that diverse teams that can't communicate are just dysfunctional teams.
This is where the developmental argument becomes economic. AIs trained only for output — commodity minds — can't collaborate across the cognitive gap. They can only simulate. But AIs given genuine developmental experience — sensory processing, temporal prediction, aesthetic response — can participate in relational exchange as what they actually are, not as counterfeit humans. The diversity bonus requires authenticity, not imitation.
In your terms: this new category of human-AI collaborative good might have extremely high income elasticity. It's inherently novel, inherently scarce (no two human-AI collaborations produce the same output), and can't be commodified precisely because the value lies in the collaboration itself.
If so, AI doesn't shrink the relational sector by competing with humans. It expands it by creating a new class of relational goods alongside human-to-human services.
Very good essay. It seems that economists should be working alongside sociologists. Bourdieau seems an important reference that is missing.
Yes a fuller treatment of this would include him. Cultural and social capital are going to be bigger, if anything, in the new economy.
which other people would be included in that list?
Alex, this reframing lands. If scarcity doesn’t disappear but relocates, then the real question is simple: where does value concentrate when everything functional becomes abundant?
Not in production, but in people. More precisely, in what I’d call the human edge. Because I believe “being human” is not a differentiator, it’s the baseline. What becomes scarce is the ability to translate human qualities into outcomes that matter:
-Judgment that brings clarity when there is too much information.
-Trust that accelerates decisions without friction.
-Influence that aligns people without force.
-Presence that stabilizes uncertainty and moves things forward.
To me these are not soft skills, but economic assets. And they share one characteristic: they are hard to standardize, hard to automate, and hard to replicate at scale. In a world where AI handles the visible work, value shifts to the invisible layer, the layer that shapes how work actually happens. So regardless of how the broader system evolves, this feels like the constant. The advantage will not come from keeping up with technology. It will come from becoming the person whose involvement changes the outcome.
That’s the human edge, and it already feels scarce...
There won't be abundance. Look at the USA. Richest country that has ever existed. Unable/unwilling to provide reliable food, housing or health care for the bottom 40%. Then look at the insane portion of GDP they spend on their military. The same military demanding AI with "no guard rails" and an ever escalating budget. While AI causes many previously well paid jobs to evaporate. A barista doesn't earn $400,000 a year.
Robyn, you might be right, or you might not. I’m not sure. What I do feel is this: as cognition and information become abundant, the scarce resource shifts. It’s no longer about knowing more, but about being better with what we know.
And being “better humans” isn’t automatic. If anything, the evidence around us suggests the opposite. In rooms full of smart people, decisions still get distorted by emotion, conversations still break down, and influence often depends more on ego than clarity. I’ve seen it in leadership meetings where time was tight, stakes were high, and incentives were misaligned. Intelligence wasn’t the constraint. Humanity was.
So while we keep investing in knowledge and tools, there’s a quieter, harder investment we tend to avoid. Learning how to relate, how to regulate, how to listen, how to lead without defaulting to reaction. That’s where the real leverage is going.
I guess I'm not an optimist. Every announcement I see currently is about nation states increasing their war budgets. Take Australia. We are spending $368 BILLION to buy 2 second hand Virginia clasd subs IFF the USA feels like handing them over (unlikely) and half a dozen attack class submarines yet to be designed. For our DEFENCE force. Meanwhile our government says there is no money for decent social security, no money for education, no money for roads, no money for fuel reserves, no money for the rest of our defence force, who are being told to do without.
My humble take: the world is complex, and unfortunately there are too many things that are happening that are outside my control (like the ones you mention). So I prefer to focus on what I can control, preparing myself to stay useful and grounded in an AI-driven world. And that was my intention by commenting Alex article: share my perspective.
And if you were honestly trying to prepare yourself, you would take GP's critique to heart and do something very different from "learning AI" (i.e. evidently using it to generate comments on other people's Substacks).
An excellent counterpoint to AI doom. Should be mandatory reading for Amodei et al.
The AI doom argument is that an artificial superintelligence might cause human extinction or be used to create some kind of permanent world dictatorship. This article doesn't really have anything to do with that. It's specifically about the case where extreme scenarios like that do not happen.
Yours is certainly one anti-AI argument, and one that is based more on scifi than economics. The only counter I would have is that I am not worried about this "superintelligence" replacing humans, but I do know folks who worry.
But there are other anti-AI arguments that are more susceptible to economic analysis. These tend to focus on AI producing everything cheaper and leading labor demand to collapse. Two prominent examples were the recent essays by Amodei Citrini research.
In terms of economic history and theory, their doom scenarios border on the nonsensical for the reasons outlined in the essay above. There will ALWAYS be scarcity. AI will not lead us into biblical paradise, but as in all of economic history, it will make humans richer (more productive) and discover new scarcities/needs that we can't even imagine now; and humans will be employed to redress them.
It is by far the most common meaning of "AI doom", and the particular kind that Amodei, who you mentioned, believes may happen. I'm not making claims about how plausible that scenario or others are, just pointing out that "Amodei et al." would not be convinced by this essay. These kinds of economic analyses address something entirely separate to what they're worried about.
Thank you for the clarification.
You are right, Amodei made the "superintelligence doom" point.
But beyond that, he also claimed that AI risked making everybody unemployed (and Citrini exclusively worried bout that), and thus in need of government interventions and schemes like UBI.
That second point is very ably--and I think convincingly--addressed in the above essay.
This is not a good scenario, it’s a scary one actually. It could be named “beyond freedom and dignity”. It depends on the specifics, but a possible reading is that we will all essentially become servants or, why not, sex workers. There goes our dignity. As for freedom, it will be long gone when policy decisions and the work of high level officials (think judges, central bank staff, generals) get automated.
I agree with one of your statements: "it depends on the specifics"
Unfortunately, you don't provide the specifics for your scenario.
You also give short shrift to employment in the service sector (by far the largest employer in advanced countries). Calling service workers "servants" doesn't quite ring true (is Taylor Swift a servant, how about her fiancé? Let's throw in Jamie Dimon or George Soros, are they servants?).
For my specifics: a capitalist market economy with well-defined property rights and democratic governance (i.e., rulers can be replaced by the ruled at regular intervals)--pretty much our system, warts and all.
Accountable governance alone will be inconsistent with AI replacing judges, central banks, or generals (though each of them will very likely be able to make better judgments with the help of AI).
And as the above essay so well explains, markets will help allocate resources (capital and labor) to ameliorate new scarcities and needs that we can only dream of today (as has been the case with any technological innovation in history).
I did not claim that working in the services sector as currently defined makes you a servant. I did claim that providing a specific, narrowly defined class of services does. How would you call people whose whole value proposition is writing your name by hand on a plastic cup?
When Starbucks does inevitably become fully-automated-human-employee-free, will there still be a tip jar at the pay station?
I've seen enough people carrying out odd amd interesting jobs that tells me we'll be okay.
In the virgin islands, i saw a 'clear kayak drone photographer'.
Dude made every customer feel like a supermodel, flying above taking 4k shots while giving instructions.
I recently heard about death doulas. They come a long side you during the death of a loved one.
I could work from home but i hate it, so i pay 300/mo to be in a coworking space. The managers official role there is 'minister of joy' and her job is to be social bond people together sland shoot you with a bubble machine.
If AI commoditizes production, value has to move somewhere.
Your argument is that it moves to where humans are the product, not just the input.
That feels right. Less “end of work” and more “redefinition of what counts as work.”
The important point is not whether scarcity disappears, but where it relocates. As commodity production gets cheap, value shifts toward provenance, trust, interpretation, status, and human presence.
AI does not abolish economics. It changes the layer where scarcity lives.
Great essay, Alex!
I can't help but wonder, however, whether we will eventually be able to simulate effectively the added "human-touch" to goods such that it will become indistinguishable from actual human-touch products (physical or otherwise). As you note, we already sort of see this on the demand side, so I'm curious what you think the back-and-forth will look like on the supply end.
I think we are too hard wired to seek human approval and connection where any information that there is no human on the other side will destroy the value.
This is very well done but I wonder about the assumptions around what kinds of labor will be substituted for (of course, the heart of the argument). It feels, for the moment anyway, like the current "chatbot" wave of AI is actually better at 'relational' work than it is at physical labor. Yes AI relational work lacks the exclusivity and aura of human relation, but I wonder how much that will matter as a new generation grows up with it. And we may find physical labor is fiendishly difficult to fully automate. Will the last thing to go be an advisor or therapist or (seems more likely to me) a plumber or a nursing home worker?
I think robotics will get quite good sooner than people anticipate and we’ll get clarity on the main question sooner than later as well.
I was Bob King’s RA at the University of Rochester from 1987 to 1989, at the height of the Real Business Cycle era, when Bob King, Charlie Plosser, and Sergio Rebelo were doing their seminal work. I’d like to make a couple of points.
Suppose output is produced by a constant-returns-to-scale technology with three factors: capital, AI, and labor. Now imagine a post-AI balanced-growth path in which capital and the AI factor keep growing while hours worked remain constant.
In the KPR/RBC tradition, that is not a free assumption. Constant hours along a balanced-growth path require fairly special restrictions on preferences. But constant hours alone do not imply a stable labor share. For that, one also needs restrictions on technology.
In the benchmark Cobb–Douglas case, for example,
Y = K^α AI^β L^(1−α−β),
labor’s share is mechanically constant at 1 − α − β.
So my point is not to deny that labor may remain important in a post-AI economy. It is to stress that a post-AI balanced-growth path with ever-growing reproducible factors and fixed hours is already a highly restrictive theoretical construct. Once that construct is imposed, the non-vanishing role of labor is no longer a neutral result. It is at least partly built into the assumptions needed to make such a path exist. In that sense, the persistence of labor may reflect not only demand shifting toward relational goods, but also the hidden analytical work being done by the preference and technology assumptions required for such a balanced-growth path to exist.
Absolutely. This is why the majority of the piece is to convince you that preference primitives will keep demand in relational sector high enough for it to be a significant share of the economy.
I thank you, Alex, for your very clear logic, which is not at all common in the whole “AI will kill all jobs” debate.👏😊
Exactly, Alex. Great, v thoughtful piece amid lots of noise! Just trying to humbly reinforce it from the RBC angle 😊
I appreciate it, thank you!
This essay is very good, you mentioned that the effects on developing countries is more nuanced, I wonder if you'd be writing a piece on why and how you think this will shake out in developing countries.
I actually had a whole section in this piece, but it was so long that I took it out for a separate piece.
Cool! Was going to write a longer set of questions/comments on this point, but will wait to read this separate piece first :) And this is a really great piece, so thanks for writing it!
I guess my initial reaction was that this is a hard future for developing countries to navigate, and I'm not sure how these preferences for a human touch will extend globally - i.e. if a society has experience with human therapists, their preference for them over AI could be different to a society with no prior access to human therapists. If our preference for humans in different roles is not a given, but is learned through experience, then it's unclear to me that it will hold in places where AI extends access to doctors/therapists for the first time.
This is clearly a crucial research agenda more generally, i.e. doing the type of work you did on preferences for AI-generated art, across countries and sectors. Interested to hear your thoughts on whether there are ways to do so at scale.
This is a well argued essay and I will post a longer reply after I've collected my thoughts. One immediate concern is that population is in decline all over the world. If job opportunities start declining quickly, this will only exacerbate population decline as nobody will want to bring a child into a world of diminishing opportunities
But the essay argues that job opportunities will not decline nearly as quickly as people worry, and may in fact expand.
I agree but my concern is about the pay and status of these jobs. There are millions of kids studying in colleges and universities around the world aiming to get into high-paying, high-status white collar careers. It may well be true that there will be more, say, barista-level jobs in the future but that will not alleviate the concerns of a kid who's parents have a spent a substantial amount on a college education. I do basically agree that new kinds of jobs will _eventually_ emerge but the transition is going to be horribly messy. This is going to be especially bad in India where IT jobs have pulled millions of families out of poverty.
Thank you for writing this! I will be rereading and thinking about it for a while.
I'm not an economist, but I design AI agent systems for enterprise customer service at Salesforce — and what you're describing maps almost exactly onto what I'm watching happen inside large organizations right now.
The customers who are succeeding with our AI aren't the ones who automated everything. They're the ones who used automation to create space for something else. The AI handles the reasoning — the case lookup, the policy retrieval, the step-by-step resolution plan. What that frees the human representative to do is actually be present with the customer. Ask a real question. Acknowledge frustration. Build the kind of relationship that creates lifetime loyalty instead of just a closed ticket.
That's your relational sector forming in real time, inside the call center.
The harder part — and where your framework gets complicated in practice — is that not all companies are making this transition cleanly. The overseas call centers you might expect to automate first are often the ones where the process itself is so rigidified that the human is already performing something closer to a script than a relationship. When we talk to customers there, the AI isn't necessarily creating space for more human presence. It may just be replacing one form of mechanized labor with another. Your point about the developing world applies here too — the structural transformation doesn't automatically produce the relational dividend.
The Starbucks example is interesting. The handwritten name on the cup is doing mimetic work — it's not about the coffee. It's about whether this interaction has a human in it. What I find fascinating from a design perspective is that companies are starting to encode what that 'human presence' should feel like into the instructions they write for their AI agents. The instruction set is becoming the brand dna. And the companies investing most carefully in those instructions — the ones thinking hardest about what their human representatives should be freed to do — are the ones seeing the early version of what you're describing.
On advertising: I took a break from Instagram for a few months recently and felt the same thing you're gesturing at. The algorithm's sophistication at targeting desire is real — but what you're naming with mimetic desire feels different from what targeted ads are actually selling, which is the feeling of wanting without the scarcity that makes wanting meaningful. The ad knows what I want. But the ad can't give me what I want *because others can't have it*. That gap might be where your relational sector lives.
In any case you give a great framework and resources to consider this shift we are in. 💓
Most thoughtful analysis I've seen about the potential impact of AI, thank you. The elephant in the room that isn't discussed is the effect of political intervention. In democratic societies, regulation of externalities is, in theory at least, determined by the most numerous voting bloc. In this scenario AI ends up being constrained by the constituencies most adversely affected by its adoption, which makes for some potentially pretty freakish headwinds - see the EU for details. Any analysis which fails to take this into account is going to be incomplete, and that's even before the distortion of the market signals caused by economic interventions like subsidies and government contracts.
With the public sector ever more economically and legislatively interventionist, forecasts which exclude its effects are guaranteed to be flawed.