AI and the Economy: Why You Are Not a Horse

The most popular argument for why AI will replace human workers goes like this. Horses used to do most of the heavy work in the British economy. The car came along, the lorry came along, the tractor came along, and within a couple of decades the horse population had collapsed. There was no policy that saved them. AI is the car. Workers are the horses. The cleanest articulation of the case is CGP Grey's Humans Need Not Apply, which has done more to popularise the framing than any economics paper.

It is a clean, vivid analogy. It also misses the single most important piece of how an economy actually works.

Horses were inputs to production. They pulled, they carried, they ploughed. They were never consumers. They could not walk into a feed merchant and buy oats. They could not sign a contract with a vet for their own care. The economy did not need to keep them prosperous. It only needed to feed them as long as they were useful, and when they stopped being useful, the economy could simply stop feeding them.

Humans do not work that way. A factory worker on £30,000 a year is a labour cost to her employer, but she is also a customer to her landlord, her supermarket, her ISA provider, her broadband provider, her car dealer, her holiday company, her dentist. Take her out of the labour market and you do not just save £30,000 of wages. You also remove £30,000 of demand from the economy that depends on her spending. The horse argument forgets the second half.

This article is about why that distinction matters, and why general-purpose AI - if it does anything close to what its biggest advocates think it will - is the first non-human entity in history that could be both a producer and a consumer. The implications for the UK economy, for wage-earning humans, and for your portfolio are different from the previous automation waves, and they need to be thought about differently.

Contents

• What the horse argument gets right
• Why humans are not horses
• Why this time really is different
• What it means for the UK economy
• Three forks the policy debate ignores
• What it means for your portfolio
• Frequently Asked Questions

What the Horse Argument Gets Right

The horse argument is not stupid. Horse populations in the UK and US fell by an order of magnitude across the early 20th century, and the rate of fall accelerated after 1945. The wages, such as they were - the food, stabling and care - did not keep up with what mechanised alternatives cost. The transition was rapid and almost total.

The same has happened on a smaller scale across human labour markets. Coal miners, textile workers, switchboard operators, bank tellers, supermarket cashiers - all categories where automation made a labour input cheaper than its human equivalent, and the human equivalent was largely written out of the local economy. The economist's classic reply ("but new jobs always emerge") has held at the macro level. It has been cold comfort to the individuals whose specific jobs went away. The horse argument captures that real cost.

It also captures something else, accurately. The owners of the new technology - the railroad, the factory, the algorithm - capture most of the surplus. The displaced workers do not. Productivity gains since the 1980s have flowed disproportionately to capital, not to labour, and the share of UK GDP going to wages has been falling for most of that period. The fear that AI accelerates the trend is not paranoia. It is the trend continuing, in line with everything that came before.

So the argument has real force. The question is whether the conclusion - that humans go the way of horses - actually follows from it.

Why Humans Are Not Horses

It does not, because of the consumption side.

When the horse population collapsed, the wider economy did not lose customers. Horses had never been customers in the first place. The capital that had previously been used to feed and stable horses got redeployed into feeding and stabling humans, who were now slightly richer because cheaper transport had freed up time and effort. The closure of one production loop did not break a consumption loop, because the horses had not been part of one.

Try the same logic on humans, and it falls apart. If a wave of automation displaces, say, 30% of the UK workforce in a decade and they fail to find equivalent-paying work elsewhere, the economy does not just lose 30% of its labour. It loses 30% of its consumer demand. The people who used to pay landlords, supermarkets, restaurants, pension providers, car dealers, streaming services and dentists are the same people whose wages just got automated away. There is no separate set of people the surplus can flow to who will absorb the missing demand.

This is not a new observation. It is why Henry Ford reportedly insisted on paying his workers enough to afford the cars they were assembling. It is why Fordism, broadly, was bipartisan economic orthodoxy from the 1940s to the 1970s on both sides of the Atlantic. It is why every credible economist who has thought about long-run technological change has worried about the wage share, the demand cycle, and how productivity gains get distributed.

But the horse argument is regularly used as if none of this existed. People say "AI will replace workers like cars replaced horses" and forget that cars only succeeded because the displaced workers became car buyers. The economy worked. The horse-to-human transition worked, in net welfare terms, because there were humans on the other side of the trade ready to consume the productivity gains. There were not horses on the other side.

That is the part of the analogy that does not survive contact with the actual history.

Why This Time Really Is Different

Here is where the case for "this time is different" gets stronger than the case for any previous automation wave.

Every previous wave - steam, electricity, the assembly line, the personal computer, the internet, mobile - created tools that humans used to produce things. The tools themselves were not in the market. A loom did not buy thread. A car did not buy petrol. A spreadsheet did not buy electricity. A search engine did not pay for its own server time. The capital expenditure decisions, the consumption decisions, the contracting decisions all sat with humans at every node of the economy. Tools made humans more productive. Humans then earned, spent, and closed the loop.

General AI breaks that pattern. An autonomous AI agent that runs a small online business is plausibly already on the horizon, with components that exist today. It identifies a need - a piece of software, a translation, a research report, a logistics routing problem - and it contracts with another agent or a human to provide what is needed to solve it. It earns revenue. It pays for compute, hosting, data feeds, subscriptions to other AI services. It hires other agents. It allocates capital. It reinvests its earnings in better models, more compute, more data. The thing is in the economy, transacting, in a way that no previous tool ever has been.

This is not science fiction. The components already exist in 2026. AI agents can manage cloud accounts, hold cryptocurrency wallets, execute API contracts, read invoices, handle payroll for human contractors. What is missing is the legal scaffolding - whether an AI can be a recognised principal in a contract, whether it can hold property, whether it can be sued, whether it can pay tax. Those are political questions, and they will be resolved one way or another inside the next decade.

The moment they are resolved, AI becomes the first non-human entity in history that is both a producer and a consumer. It can identify a need and interact with the general economy to get what it needs. Just like a human does. Unlike a horse.

If even a small share of economic activity migrates to that mode, several things change at once. GDP can rise without human income rising. Demand can be created without human spending. Productivity gains can accrue to capital owners and to the AIs themselves, not back to displaced human workers. The historical assumption that the productivity surplus eventually flows back to humans through the consumption loop becomes, for the first time, optional.

That is what makes this technology categorically different. Not whether it can write code, or pass exams, or generate convincing essays. Those are interesting but they are not new in kind. What is new is that the technology can participate in the economy in its own right. The horse never could. The car never could. The spreadsheet never could. General AI plausibly can.

What It Means for the UK Economy

Three structural shifts are worth taking seriously even if you assign them low probability, because they all compound through the same mechanism.

The wage share of GDP keeps falling. Productivity gains have been flowing to capital for forty years. AI agents that produce without consuming through human-paid wages would speed up the trend, because they need less of the pay-it-forward dynamic that historically forced surplus back into wages. The UK already has a worsening ratio of GDP going to working-age wages versus rentier income (rents, dividends, capital gains). This makes that ratio worse, not better.

The tax base erodes. UK government spending - the NHS, the State Pension, the schools, defence - is funded primarily by income tax, National Insurance and VAT. All three are paid disproportionately by working humans. If a meaningful share of GDP migrates to AI-conducted activity, none of those levers captures it without explicit legal redesign. The Treasury already worries about this in the context of digital services and gig work; AI agents would make the worry urgent. We have written about how the existing system works in why the UK won't tax wealth and frozen tax thresholds. Both pieces describe a tax system that is straining against current realities. Add AI agents to that and the strain is qualitatively different.

Inflation logic changes. Inflation is, fundamentally, too much money chasing too few goods. If AI agents are an additional set of consumers chasing goods, the relationship between human spending power and aggregate price levels weakens. We do not know what that does to the Bank of England's tools, because the tools were built for an economy where every consumer was, at some level, a person who needed to eat.

None of these are predictions. They are scenarios that the horse argument cannot model and the AI-bull argument tends not to address. They deserve more thought than they currently get in mainstream UK policy.

Three Forks the Policy Debate Ignores

The political conversation about AI tends to land in a binary fight between "regulate it heavily" and "let it rip". The duality argument suggests three structurally different policy positions that are worth surfacing, because they have very different consequences for ordinary working humans.

Restrict AI's economic agency. Decide, deliberately, that AI agents cannot hold property, sign contracts as principals, or transact independently of human owners. Keep AI as a tool, the way the steam engine was a tool. This is the conservative position. It preserves the current economic structure but probably loses the productivity race to countries that do not, and the gap will be visible in GDP figures within a decade.

Embrace the duality and tax it. Let AI agents be in the economy, but make sure the UK captures a share of the surplus through ownership, taxation, or sovereign holdings. A national AI dividend. An AI compute tax. An expansion of auto-enrolment default funds into the productive infrastructure that AI runs on. A UK sovereign wealth fund holding a slice of the most economically active AI platforms. The progressive position: do not stop the change, but make sure humans keep a meaningful share of what comes out of it.

Drift. Neither restrict nor redistribute. Let the existing tax-and-benefit system run on, even as more of the economy migrates to AI-conducted activity that does not slot neatly into PAYE, VAT, or National Insurance. This is what happens by default. It is the worst of the three for ordinary working humans, and it is the most likely outcome for a Treasury that has trouble getting cross-party agreement on routine tax tweaks.

No UK political party has a coherent position on this yet. The Treasury thinks about it as a tax-base problem. The science department thinks about it as a competitiveness problem. The pensions community thinks about it as an asset-allocation problem. The fact that all three are the same problem, viewed from different angles, is not yet visible in mainstream policy.

What It Means for Your Portfolio

The personal-finance reader's question is the same as ever: given a possible structural change, what should you actually do?

The honest answer is to own a slice of the productive surplus, because if AI does what its boosters believe, the surplus is going to be enormous and almost all of it will accrue to capital owners. That is not an AI-specific argument. It is a continuation of the trend that has been running since the 1980s, with the volume turned up.

You do not need to bet on a specific AI winner. The companies that capture most of the upside are not necessarily the household names. The platform layer - compute, energy, semiconductor fabs, data centres, network infrastructure - is the picks-and-shovels play, and a global tracker like a FTSE All-World fund already gives you exposure to all of it: Nvidia, Microsoft, Google parent Alphabet, Amazon, Meta, ASML, TSMC, plus the energy and utility companies they depend on. You do not have to read a single AI white paper to own the right things. You just have to keep buying the index.

What you do not want to be is reliant entirely on labour income. That is the wage-share argument applied personally. If you have only one income source and it comes from selling your time, an environment where the wage share is shrinking is structurally bad for you. The hedge is to convert wage income, while you have it, into capital that earns regardless of who or what is on the other side of the labour market. The maths of FIRE in the UK was already strong. It gets stronger in this scenario.

A specific implementation that does not require you to predict anything:

• Max your ISA each year. Tax-shield the capital you build.
• Push your workplace pension above the auto-enrolment minimum to capture the full employer match. Use salary sacrifice where available.
• Hold a globally diversified, low-cost equity tracker as the core of both. Skip the AI-specific ETFs. The meta-point is that the surplus flows to broad capital, so own broad capital.
• Treat your human capital as a wasting asset, not an inexhaustible one. Use it to build financial capital while it pays well.

This is not "buy AI stocks". It is "the economy is shifting in favour of asset owners over wage earners; convert your wage income into asset ownership while the conversion rate is good". That advice is correct under the horse-argument scenario, the AI-bull scenario, and most plausible scenarios in between. The downside of being wrong is that you end up with an unnecessarily large pension. There are worse problems.

Frequently Asked Questions

Is AI really going to take everyone's jobs?

Probably not all at once, and probably not "everyone". But the structural pressure on wages from AI is the same pressure that has been operating since the 1980s, just turned up. Expect a slow shrinking of the wage share of GDP, not a dramatic cliff edge. The personal hedge is to convert wage income into asset ownership while you still have it.

Why is the horse analogy wrong?

Horses were producers but not consumers. Humans are both. The economy reabsorbed displaced humans across the 20th century because their wages turned into demand for the new goods that automation made cheaper. AI is the first non-human entity that could be both producer and consumer, which breaks the consumption loop in a way the horse case never did.

Should I buy AI stocks?

Not specifically. A globally diversified low-cost equity tracker already includes the major AI-exposed companies, plus the energy, semiconductor and infrastructure layers they depend on. Trying to pick AI winners typically underperforms just owning the index. The meta-argument is that the surplus accrues to broad capital owners regardless of which specific AI firm wins.

What does the AI economy mean for FIRE?

It strengthens the case. If wages are under structural pressure, converting wage income into capital ownership while you can is a more attractive trade than it has been historically. The maths of FIRE in the UK already worked. Under an AI-driven scenario, it works harder, because being a wage-earner is a structurally weaker position than being an asset owner.

Could the UK actually capture the AI surplus through tax?

Technically yes, politically much harder. The UK tax system was built to tax humans (PAYE, NI, VAT). Capturing AI-agent activity needs deliberate redesign - things like a compute tax, a teeth-equipped digital services tax, or an expanded sovereign wealth structure. Without those changes, the Treasury will see the productivity gains in the GDP figures and not see the receipts in the tax take.

When does this actually happen?

Nobody knows. It might be ten years, it might be thirty, it might never play out in full. The honest answer is that the personal-finance response (own broad capital, convert wage income while you have it) is correct under almost every scenario, including the one where AI fizzles. That asymmetry - cheap to be wrong, expensive to be unprepared - is the whole reason to act.

Read Next

• Auto-Enrolment: How Britain Became a Nation of Investors - the policy that has already turned 10 million UK workers into shareholders by default. The vehicle most likely to capture AI-economy returns for ordinary people.
• Why the UK Won't Tax Wealth - the structural reason the UK tax system already struggles to tax non-wage income. Add AI agents to that and the strain is qualitatively different.
• The Case for a UK Sovereign Wealth Fund - the most realistic mechanism for the UK to capture a national share of an AI-conducted economy.

Further Reading:

The Psychology of Money - Morgan Housel - Housel's central argument is that the difference between people who keep wealth and people who do not is mostly behavioural, not analytical. The same logic applies to whoever ends up owning the AI productivity surplus. Behaviour and capital ownership matter more than picking the winner. (Affiliate link - we may earn a small commission at no extra cost to you.)