Automation: Is This Time So Different, Really?

Ladies and Gentlemen, thank you for tuning in. To kick off this deep dive, I’d like to start with an illustrative tweet by an obscure figure: the man formerly known as Elon Musk, now full‑time culture warrior and divisive figure. It’s fair to say Robo‑topia (and Dystopia) is having a cultural moment. 

Putting aside the obviously alarming question of who owns the means of production in a 100% robot economy- and how we buy robots without money from our jobs(??)- transformative technology is now part of the zeitgeist. And yet, it is nothing new. Neither is the moral panic about the future of work, humanity and society. Since the first industrial revolution, the core question we have always asked is: “This time, will jobs destroyed be more than jobs made? Will this time be different?”

In short, we have been perennially predicting the end of the system as we know it.​

But, alas, as we all know, the end of society is not what has unfolded. Despite having a very, very automated economy compared to even 50 years ago, jobs have always increased at a faster rate than they were destroyed. As far as we can tell, the picture currently being painted is the same: jobs created are still predicted to be net positive, not negative.​

Yet people are, like time past, saying this time is different. They are right, but also wrong.​

Now, this article could age very badly, but it seems the data isn’t saying yet that the technology is radically more transformative than the previous waves, at least in job and productivity measures. But my point is that it doesn’t have to be for us to be worried about it- and currently by focusing on long term end states (will robots do everything) and aggregate figures like net job creation, we are missing the immediate risks here. Technology has and will always reallocate work and change the economic system. These transitions are real and painful, sometimes even bloody. 

Instead, we should be thinking not about the long run, but how we get there, and the challenges this brings. What is the context in which this automation is taking place? Do the big trends of the last 40 years- namely ageing, weaker dynamism, inequality, and fraying institutions- mean the system has slack to adjust to the changes ahead? If the “buffers” that allowed society to absorb the last big manufacturing shock, those of trust, mobility, fiscal space, and time, are fragile and arguably gone- which ones are we calling upon now?​

So rather than asking again, “Will there be any jobs left?”, a more pressing question is: what does the transition look like this time, and do we actually have the capacity to manage it?

Automation has an anti-memetic quality, in that we forget just how much technology reshapes our lives: partly because (1) we weren't there riding the horse and cart in 1900, and partly (2) because our brains adapt very, very quickly.

In 1900, around 41% of the US workforce was in agriculture, and by 2000 it was roughly 2%- and we make far more food for far more people with that 2%. So we automated nearly half the economy out of existence, yet unemployment mostly hovered around 4-5% while the population exploded, which raises an obvious question about how that works. The answer is that previous technological revolutions didn't just create jobs as populations grew; they enabled the population growth that needed those jobs in the first place.

For most of human history whenever a society got a productivity windfall e.g. a good harvest from better ploughs, more children survived and the population grew, but then eventually the extra mouths ate through the surplus. So living standards drifted back towards subsistence and the cycle started again, which is what economists call the Malthusian Trap. The graph below shows this: nearly three thousand years of income per person going basically nowhere, and then around the Industrial Revolution, productivity starts compounding faster than the population can absorb it, living standards pull away from subsistence, and that pattern breaks; what historians call the Great Divergence, visible as that sudden upward kink in the graph.

The mechanism of automation leading to more, not less jobs is a virtuous cycle that goes something like: when automation makes production cheaper, companies can lower their prices. That means everyday people have more money left over to spend on other things. That new spending creates demand for products and services that didn't really exist before and someone has to make, deliver, and support all of it. So rather than eliminating jobs overall, automation tends to shift where the jobs are, and often creates more of them in the process. That doesn’t mean people don’t lose their jobs- they do. It’s just that new jobs get created elsewhere, and the economy shifts in its structure. That is how we moved from agrarian, to manufacturing to ultimately a services based economy in advanced industrialised nations. 

Today's AI projections mostly assume this flywheel still works. The World Economic Forum expects roughly 92 million jobs to be displaced and about 170 million created by 2030. So, yes we are seeing jobs being lost due to AI, in fact just this morning I saw Jack Dorsey is letting go 4,000 of his 10,000 staff at Block. But the idea is that new jobs will be created elsewhere from these increased efficiencies eventually (although that's little comfort to people either losing their jobs or terrified they will be in the near future).

Nonetheless, a lot of people in tech argue in specific and compelling ways that AI's general-purpose, cognitive nature will blow up this virtuous cycle. I don’t have time to cover these here but I encourage you to check out this article by Lawrence Lundy Bryan and the great economics blog, Noahopinion, which opines extensively on what the data currently shows. Now, it almost goes without saying, economists are famously bad at predicting regime shifts. Which makes sense: when the future is working so differently from the past, historical data just isn’t helpful to explain these new futures. 

And yet I still think it's the wrong immediate focus, or at the very least it's getting too much airtime imo. In this framing, the eventual net job number is the de-facto collective risk indicator/existential threat monitor of AI because the nightmare scenario we're imagining is a world without work, when the more immediate danger is probably in the roughness of the transition itself: how many people are forced to move, how far they fall, how fast it happens, and how this fallout reverberates onto everyone else. Can our institutions absorb that shock? It doesn't have to swing to a net jobs negative scenario for that to be a concern. And it’s this cheery avenue I’d like to explore. 

History shows both the effects of this strain from reallocation and how misleading the aggregates/end state focuses can be when it comes to the labour market. Studies on the China trade shock give a more recent and clearer illustration. At the national level, the US through the 1990s and 2000s looked fine: GDP was growing and unemployment mostly sat between 4% and 6%. But in the commuting zones most exposed to Chinese import competition, Autor, Dorn and Hanson find something very different. 

Manufacturing jobs disappeared and didn’t come back locally; employment-population ratios dropped; wages and household incomes stagnated or fell relative to less-exposed regions; unemployment and non-employment stayed elevated for at least a decade. Those places also saw higher disability claims, more poverty, worse health outcomes, and a lasting shift towards more anti-establishment political candidates. In the aggregate, those losses are offset by gains elsewhere; cheaper goods from China raise the disposable income of others and drive more jobs in services (both low and high-paid ones), so the “net” effect looks manageable. 

In lived experience, a small set of regions absorb most of the damage and never really get back to their previous trajectory. Take another example, Britain’s former coalfield areas (e.g County Durham): there are only 57 employee jobs per 100 working‑age residents, compared with a national average of 73 and 88 in major regional cities, even decades after the pits closed. 

The aforementioned studies are, of course, trade shock studies rather than automation shock ones- but automation has the same end effect in that it creates sticky and long-term labour market disruption, albeit through a different mechanism to offshoring. 

Economists tend to think about jobs as bundles of tasks, and automation doesn't hit jobs but tasks, unbundling them and rebundling them into new configurations and shifting the labour market as a result- and workers can only switch into new jobs to the extent they are geographically available to them and they have the required skill set, which is not a given. This is why we see long-lasting damage to certain areas and people (think Rust Belt) who bear the bulk of the cost in ways that get hidden by aggregate statistics. This is also the primary mechanism through which technology drives income inequality, with one recent estimate suggesting automation alone explains more than half of the rise in the wage gap between more and less-educated workers in the US over the last four decades.

Now add AI to that picture. Moving US labour out of agriculture took most of a century, the China shock played out over 10–20 years and still left deep scars in the regions it affected. The current wave of AI tools can (and has started to) change the task mix of a broad base of people previously thought to be immune to this kind of automation: lawyers, investment bankers, customer-support teams and analysts within just a couple of years. Graduate hiring in many white-collar tracks has already been cut back sharply, with surveys in the US and Europe reporting double-digit percentage drops in entry-level openings in fields like consulting, finance and tech since the post-pandemic peak, something that many are attributing largely (but not solely mind you) to AI. 

Since skills systems, local economies and political institutions simply don't adapt that quickly, the strain of transition, and particularly the timespan of said transition means relatively modest but rapid shocks concentrated in specific sectors, peoples and places is enough for things to go badly without any sci-fi scenario being required. Think 10% more regional and 10% more youth unemployment than ever before in 5 years not over 20 years, and certainly not the economy-wide, 50% unemployment and net job destruction scenario imagined by the robo-topists (although this world is clearly far worse).

Now, this is nothing new- we’ve been going through these cycles for a long time and the optimistic story that economies ultimately adapt and new industries emerge is definitionally right (this is survivorship bias- we all live in societies that have lived to tell the tale). But it glosses over how much institutional input previous transitions involved and the timespans taken. Past waves of big technological change were not “markets did their thing and everything came out fine”; technology, markets, social norms and institutions ultimately co-evolve in some way, shape or form over time.

Take Engels' Pause as an example- or a good example of what happens when institutions arguably don’t intervene for a while. Between about 1780 and 1840, British output per worker rose roughly 39-46% while real wages rose only around 12%, so for roughly fifty years most of the gains went to capital and higher earners and working-class living standards were basically flat despite rapid growth. The period was also marked by frequent industrial disputes, food riots and machine-breaking protests, which many* historians interpret as responses to exactly that dislocation and lack of redistribution.

Ultimately, the move from agrarian to industrial economies went hand in hand with the expansion of mass education, with literacy and basic numeracy rising across the 19th and early 20th centuries as compulsory schooling laws spread and secondary education became more accessible (this happened I should add, for many, widely debated reasons). But fortuitously, it allowed for more to switch to clerical work from agrarian (and for more growth to happen as a result of this institutional meddling, not less). 

 In the mid-20th century, as mechanisation and mass production spread, many advanced democracies built or expanded social-insurance systems, strengthened labour protections, and undertook large-scale public investment in infrastructure and housing. After World War II, globalisation was accompanied by the construction and extension of welfare states in Europe and large federal programmes in the US like the GI Bill and the interstate highway system.

The macro backdrop of the post-war era made this sort of institution‑building unusually tractable (and yet as we all know, by the 1980s a regime change still came). In countries like the US and UK, debt‑to‑GDP ratios fell sharply from wartime peaks as growth outpaced interest costs whilst both the welfare state and working‑age populations were relatively young. Baby boomers and their rising real incomes dramatically expanded the tax base. In the US, trust in the federal government peaked above 70% in the mid‑1960s and remained high through much of that decade, which made it easier to pass and sustain ambitious programmes. These institutions had fiscal space, demographic tailwinds and political legitimacy

There has always been friction between the slow-moving institution and the fast-paced market. AI arguably 10x-es this due to its pace, and with it, that friction.

Today on the market side things look energetic; capital is pouring in, bright young things are taking on massive challenges at the frontiers of science and what's possible arguably more than ever (an arena that I feel lucky to have a front row seat to everyday), AI adoption is spreading across sectors, firms are reorganising around AI capabilities. Survey data show that frequent AI use at work has nearly doubled in the last two years, driven mainly by white-collar roles. AI has a faster uptake rate than any earlier general-purpose technology at comparable points.

On the institutional side however, the picture is quite different- one almost inverted relative to that post-war picture. Today a much larger share of public spending is now pre-committed to mandatory programmes; such as pensions, healthcare, social care (aka lots of essential stuff that supports an ageing population) as well as interest payments, leaving much less room for governments to shift spending at the margin. E.g. US federal debt held by the public has climbed from roughly 30% of GDP around 1980 to something like the 100% range today, with gross debt near 120%, which makes for e.g. large, debt-financed transition programmes a harder macro and political sell if we were to decide that's what's required. Trust in the federal government now hovers below 20–30% in consistent polling. 

Many advanced economies show similar combinations of higher debt, ageing populations, heavier entitlement burdens and declining trust. Meanwhile, AI is not confined to a narrow band of work; it is already being applied across a wide range of work at an unprecedented pace for technological diffusion, and with it social unrest and instability, if you extrapolate out the findings from history (or even the ones mentioned in this article). 

Strained institutions deliver worse outcomes, which erodes trust, which makes reform harder, which strains institutions further. AI doesn’t have to be historically unique to land in the middle of that and make it significantly worse.

I write this from a world that is, mostly rightly, sceptical of governments, and from a sector that, mostly rightly, values markets and individual agency. But I think we're kidding ourselves if we treat market failures as edge cases. The transition to AI-native everything will create significant market failures that will need addressing. 

I'll be direct about why this matters to me: I'm increasingly worried about democracy and with recent fertility hysteria and abortion rollbacks, about my rights as a woman. I don't think it takes a genius to recognise this is related to what's happening in labour markets. I worry that political instability, then democratic and institutional erosion, will come and with it a shrinking space for the kind of pluralism that protects everyone.

So the question I’d leave you with isn’t “will there be enough jobs?”: it’s whether we’re willing to do the hard, slow, unglamorous work of keeping institutions functional enough to manage what’s coming. That means engaging with policy, with politics, with the machinery of collective decision-making- even when it feels broken, perhaps especially then. Markets are extraordinary. They are not sufficient. In the window between now and whatever equilibrium we land in, what happens to real people in real places will be determined not just by technology but also in large part by whether we show up for the institutional work. That feels, to me, like a reason not to check out.