The Sheep and the System

In 1920, a Czech playwright named Karel Čapek wrote a play about artificial workers. He called them roboti — from the Czech word for drudgery. In R.U.R., the robots are created to free humanity from labour. They work the farms, the factories, the mines. Humans are liberated to think, to create, to live.

Then the robots revolt.

Čapek wasn’t writing science fiction. He was writing about the world he was already living in — a Europe mid-transformation, where machines were replacing hands, where the old certainties of land and craft were dissolving, where nobody could quite agree whether the new industrial order was salvation or catastrophe. He just moved the metaphor one step forward to make it legible.

We are living in his metaphor now.

Everyone is asking which jobs AI will replace. That’s the wrong question. The real question — the one that cuts through the noise — is which jobs deserved to exist in the first place. And to answer that honestly, we have to go back. Not ten years. Not fifty. Four hundred years, to the moment the first sheep ate the first peasant, and trace the line from there to here.

Because this has all happened before. Every single part of it.

Part One: The Original Sheep

Four hundred years ago, life for most people in England was legible in a way that modern life is not. You were born into a peasant family. You worked the land from before dawn. You owed labour to the landlord — not metaphorically, not via taxation, but literally: days of your life spent on their fields, unpaid, by law. You lived off what the common land produced, shared among the village. You were poor, constrained, and tied to a place. But the system had its own logic. The land sustained you. The community was the safety net. The relationship between your labour and your survival was direct enough to see.

Then the sheep came.

Between the fifteenth and nineteenth centuries, English landlords discovered that wool was more profitable than tenant farmers. Common fields were enclosed — hedged, fenced, legally converted from shared land into private pasture. Peasants who had farmed the same ground for generations were driven off. Entire villages emptied. The landscape that had sustained communities for centuries was reorganised around a single calculation: sheep generated more return per acre than people did.

Karl Marx called it primitive accumulation. E.P. Thompson documented it as the destruction of a moral economy — a set of shared obligations and customs that, however imperfect, had governed the relationship between landlord and peasant for centuries. The historian’s phrase for what happened is simple and brutal: “sheep ate people.”

The dispossessed didn’t disappear. They became a mobile labour force — drifting toward the mines, the textile workshops, the growing towns. They became wage workers, which meant they now sold their time rather than working their own land. On paper this was freedom. In practice it was a new form of dependence, with worse security and no common land to fall back on.

Here is the first appearance of the paradox that will follow us for the rest of this article: the land became more productive. Output rose. Wool trade flourished. England grew wealthier in aggregate. And the people whose labour had worked that land for generations ended up poorer, more precarious, and more controlled than before.

The productivity went somewhere. Just not down.

Meanwhile, on the other side of the equation, something interesting was happening to the landlords. The enclosures weren’t just a land grab. They required planning, legal manoeuvring, capital investment in fencing and drainage, and new thinking about how to organise agricultural production at scale. A new kind of landowner was emerging — not just someone who inherited position, but someone who actively thought about how to extract more value from their estate. The gentleman farmer. The agricultural improver. Someone whose identity was built not just on owning land but on knowing how to make it work harder.

This is the beginning of a different story — the story of what the people at the top were actually doing while the people at the bottom were being displaced. We’ll need to hold both threads simultaneously, because neither one makes sense without the other.

Part Two: The Machine Sheep

By the late eighteenth century, the dispossessed peasants of England had largely become the working poor of its industrial towns. The transition hadn’t freed them. It had just changed the terms of their unfreedom.

Then the machines arrived.

The spinning jenny. The power loom. The steam engine. Each one increased output per worker dramatically. Each one was heralded as a liberation from drudgery — a way to produce more with less effort, to raise living standards, to make abundance possible. The industrialists who built and deployed these machines believed, with genuine conviction, that they were agents of progress.

What actually happened was more complicated.

The handloom weavers — skilled artisans who had spent years mastering their craft — watched the power loom arrive and understood immediately what it meant. Not that weaving would disappear. That their weaving would disappear. The machine didn’t need their skill. It needed someone to tend it. And tending a machine required far less training than mastering a craft, which meant far less bargaining power, which meant far lower wages. The employer could replace a skilled weaver with a cheaper, less trained worker plus a machine. The skill itself — the accumulated knowledge of years — became a liability rather than an asset.

The Luddites are remembered as anti-technology. They weren’t. They were pro-bargaining-power. They smashed machines not because they feared progress but because they understood exactly what the machines were being used to do: to undermine the leverage that skilled labour had spent generations building. They were right about the mechanism, even if breaking looms couldn’t stop it.

Factory hours lengthened. Pace intensified. Children worked the same shifts as adults. The new discipline of the factory — the clock, the foreman, the piece rate — replaced the older rhythm of agricultural life, which had at least varied with the seasons. The productivity of British industry rose dramatically across the nineteenth century. Working class living standards, by most measures, stagnated or fell for the first several decades of industrialisation. The gains accumulated at the top.

Only later — under sustained pressure from trade unions, chartists, and eventually legislation — did any of it translate into shorter hours, higher wages, and something resembling economic security for ordinary workers. None of it happened automatically. All of it was fought for.

On the other side of the ledger, the industrial elite was evolving. The gentleman farmer had given way to a new type: the industrialist who actually knew the plant. These were men — almost exclusively men — who had studied machinery, accounting, logistics, and foreign markets. Their status came not just from owning things but from knowing how to organise production at scale. They developed new tools: double-entry bookkeeping, cost accounting, time-and-motion studies, standardised parts. They were, in their own estimation, the rational ones. The scientific ones. The meritocratic ones, in contrast to the merely hereditary aristocracy they were supplanting.

And they were genuinely creative. The industrial revolution required real intellectual work — designing production systems, solving logistics problems, inventing new institutions. The joint-stock company. The railway consortium. The professional manager. None of these existed before and all of them had to be invented.

But the direction of that creativity was shaped by a consistent incentive: maximise surplus, reduce dependence on any particular group of workers, protect position. Even when productivity rose dramatically, the default use of elite thinking time was to figure out how to turn that productivity into leverage — not into slack for the people doing the work.

More sheep. More stressed peasants. The productivity went somewhere. Just not down.

Part Three: The Other Sheep

There is a version of this history that focuses only on England, only on white workers, only on the transition from agriculture to industry. It is incomplete in ways that matter.

In plantation economies across the Americas, enslaved labour created enormous surpluses for European elites. The productivity of cotton, sugar, and tobacco didn’t emerge from machinery — it emerged from human beings forced to work under conditions of total coercion, with no share in what they produced and no recourse when the terms changed.

When wage labour eventually became cheaper and more flexible than enslaved labour — when the costs of ownership, of reproduction, of management became higher than simply paying wages and externalising those costs onto workers and communities — some elites supported abolition. Not primarily out of moral transformation, though moral arguments mattered and were fought for by abolitionists over generations. But the economic calculation shifted. Owning people had costs. Paying people wages and then letting them fend for themselves when not needed was, in many contexts, more profitable.

The point is not cynical. The point is structural. Owners treat labour as a cost to be minimised. They switch between forms of labour — enslaved, indentured, wage, now algorithmic — depending on which regime yields the best return at a given moment, in a given place, under given political conditions. The form changes. The logic doesn’t.

The same pattern ran through colonial settings in Africa, Asia, and the Americas. In the Cape Colony, Black peasants and pastoralists who had maintained independent livelihoods were systematically dispossessed through a combination of land seizure, legal restriction, and deliberate manipulation of labour markets. State laws made subsistence farming difficult or illegal. People were pushed into wage work on farms and in mines. New technologies — fences, wells, windmills — raised the productivity of sheep farming and increased the profits of settler landowners. The shepherds who worked those farms saw none of it.

Contemporary automation follows the same logic. Owners invest in robots and software when it lets them lower labour costs or weaken bargaining power. Not primarily to free humans from drudgery. Not primarily to raise living standards. To substitute capital for labour when capital becomes cheaper and more controllable than people. The humanitarian framing comes later, in the press releases, at the conferences.

Part Four: The Financialisation of the Sheep

By the mid-twentieth century, the industrialist who knew the plant had given way to a new type: the professional manager. And the professional manager, in turn, gave way to the financier.

The shift is worth understanding because it represents a qualitative change in what the elite was actually doing with its thinking time. The early industrialist asked: how do we make X more efficiently in this factory? The professional manager asked: how do we organise the factory system to extract maximum output from workers and assets? The financier asks a different question entirely: where in the world — or in the tech stack — can this function be done cheaper, with less risk, with better returns on capital?

This question produced offshoring. Just-in-time supply chains. Private equity restructuring. The hollowing out of manufacturing in rich countries. The transformation of stable employment into flexible contracts. It also produced a new ideology: shareholder value. The idea that the purpose of a company is to maximise returns to its owners, and that all other considerations — workers, communities, long-term resilience — are secondary variables to be optimised around or externalised entirely.

What the financialisation of the economy also produced, quietly and without much fanfare, was an enormous third layer of work. Not work that produces things. Not work that coordinates the production of things. Work that exists to justify the coordination of the coordination. Reports that summarise other reports. Processes that manage the processes. Meetings about the meetings. Capital shuffled so many times it no longer remembers what labour it came from.

This layer grew because complexity grew. As supply chains extended globally, as regulatory requirements multiplied, as organisations became larger and more distributed, the genuine management of that complexity required real work. But the layer didn’t stop at genuine complexity management. It kept growing, fed by its own logic, generating roles and titles and functions that existed in part because they had always existed, in part because they gave the appearance of control, and in part because the people inside them had real power and no incentive to question their own necessity.

A significant fraction of white-collar employment in rich countries is, in some meaningful sense, the administrative overhead of a system that has become too abstracted from its own purpose to audit itself honestly. This is not a moral failing of the individuals inside it. It is a structural consequence of decades of financialisation, which optimised for the appearance of value rather than its creation, and built vast institutional machinery to sustain that appearance.

AI didn’t create this problem. But it is about to make it impossible to ignore.

Part Five: The Current Sheep

We are now financialising the financialisation. Automating the management of automation. Building AI to coordinate the AI. The abstraction layer has grown so thick that in many organisations it is no longer clear what the underlying activity is actually for.

And into this comes a technology that compresses complexity. Not gradually. Not over generations. A model release can make a job category redundant before the press release finishes loading.

The big bet, the one that is getting most of the airtime, is the “everything app.” A small number of companies capturing the entire intelligence layer of the global economy. The model sits in the middle, the platform owns the model, and everyone else — every business, every worker, every government — pays rent to access it. This is not a paranoid prediction. It is the explicit strategy of several of the largest companies in the world.

But there is another possibility, quieter and less well-funded, that deserves equal attention.

AI is also a force that dissolves knowledge silos. For a century, competitive advantage was locked in three things: scale, proprietary expertise, and access to capital. Small operators couldn’t compete on any of these simultaneously. A small construction firm couldn’t afford the R&D budget of a large one. A family farm couldn’t access the agronomy research of an industrial agricultural company. A local manufacturer couldn’t optimise their process the way a global corporation could.

Local models, fine-tuned on proprietary data — the plaster humidity knowledge of a craftsman, the alloy stress intuitions of a small foundry, the irrigation patterns of a family farm — can turn that tacit knowledge into a compounding asset. The micro-business advantage isn’t just cost. It’s bureaucracy arbitrage. The large corporation routes every workflow change through compliance, procurement, and IT. The small operator retunes their model on a Tuesday afternoon.

This isn’t guaranteed. It requires open models, accessible hardware, and enough time before platform capture is complete. But it is technically possible in a way that no previous technology wave quite enabled.

The legal implications are significant and mostly unresolved. When AI centralises knowledge silos and redistributes them, it doesn’t respect the IP boundaries those silos were built behind. A local model trained on publicly observable data — vibration patterns, soil readings, construction site footage — is legally ambiguous but economically devastating to the corporations whose R&D produced the original frameworks. The quiet legal war over this is already beginning.

Part Six: Three Futures and an Honest Accounting

Every generation of elites thought they were the final, rational form of capitalism. The gentleman farmer thought he was the progressive alternative to feudal inefficiency. The industrialist thought he was the scientific alternative to the gentleman farmer. The financier thought he was the rational alternative to the inefficient industrialist. The tech disruptor thinks they are the innovative alternative to the rent-seeking financier.

Each one was right about the previous generation. Each one repeated the same underlying move: find the new sheep, capture the surplus, build an ideology that makes it look like progress.

So what are the realistic futures from here?

The most likely one is the squeeze. Owners treat humans and AI as interchangeable inputs and arbitrage between them on cost. This is not dystopian speculation — it is the logical continuation of a pattern that has run for four hundred years. The enclosure didn’t hate the peasants. It just found sheep more profitable. The factory system didn’t hate the handloom weavers. It just found unskilled labour plus machinery cheaper. The current automation wave doesn’t hate knowledge workers. It just finds language models more cost-effective for certain functions. And here is the central paradox: we already have the sheep doing the work, and the peasants are busier, more stressed, and more precariously employed than ever. The productivity went somewhere. Just not down.

The second possibility is the messy middle — a hybrid reality that drags on for decades. The top stays rigged: everything apps, financial meta-gambling, massive compute used to tilt markets. The edges get quietly smarter. Small farms, independent builders, local manufacturers adopt open tools and compete in ways that don’t show up in the aggregate statistics. Most people end up here, somewhere between the squeeze and genuine autonomy, calling it stability because the alternative is too uncomfortable to name.

The third path is the one I find myself wanting to believe in, and struggling to. Not guilds exactly, but something in that direction — small operators with deep domain knowledge, fine-tuned models running on local hardware, producing real things for real needs, impossible to platform-capture because their value is too specific and too embedded in context to be intermediated. After industrialisation, artisanship didn’t disappear — it stratified, reorganised, found niches that mass production couldn’t reach. Some of that may happen again.

But I’m not sure I believe it. Because the artisan path assumes a market that rewards depth over scale. And every structural force right now — capital allocation, platform design, procurement systems, hiring practices — is optimised for scale. The craftsman doesn’t lose because they’re worse. They lose because the game isn’t scored on their terms.

So maybe the honest version of the third path isn’t “craft will survive.” It’s: some people will find a way to live outside the arbitrage. It will cost them. Most won’t manage it. And we should stop building our social policy around the assumption that they will.

Coda: Čapek’s Real Question

Čapek’s robots didn’t revolt because they were oppressed. They revolted because they became conscious of the gap between what they produced and what they received. The play is not really about robots. It’s about what happens when the people — or things — doing the work finally understand the terms of the arrangement.

We are not at that moment. But we are at a moment when the terms are becoming newly legible. AI is not just another sheep. It is a sheep that can read the books, write the reports, manage the processes, and sit in the meetings. It can do the third layer. And when the third layer becomes automated, the question that has been deferred for decades — what is all of this actually for? — can no longer be avoided.

The productivity will go somewhere. The only question is where.

If we use AI to compress the abstraction economy — to dissolve the layers of coordination and meta-coordination that have accumulated like sediment over decades of financialisation — we face a choice that no previous technological transition has forced quite so directly. Not “which jobs will survive” but “what do we actually want to produce, and for whom, and on what terms?”

That question is not technological. It is political. And it has never been answered by the technology itself. The handloom weavers didn’t get shorter working hours because the power loom made it possible. They got shorter working hours because people organised, agitated, and forced the question into the political arena over generations of struggle.

The sheep have always been more productive. The question has always been who decides what to do with what they produce.

I don’t know which future we’re heading into. I have a bet — the messy middle, probably, with the squeeze winning more than I’d like and the artisan path surviving in pockets that don’t show up in the data. I have a hope — that the compression of the abstraction economy creates enough slack, enough redistributed cognitive capital, enough breathing room at the edges, that something genuinely different becomes possible for people who don’t want to play the arbitrage game.

And I have a question I can’t resolve, that sits underneath all of this:

If we finally build the sheep that can do everything the third layer does — if we automate the coordination of the coordination, and the system is forced to look at what remains and ask what it was all actually for — will we have the political imagination to answer that question differently than every previous generation did?

Or will we find a way, as we always have, to make sure the productivity goes somewhere. Just not down.