The AI Shift Is Not Really About Tools. It is About Shifting Work From Labour to Capital

For most of human history, if you wanted more output, you needed more people.

More hands. More hours. More labour. More output. More revenue. More profit.

Similar margins. Similar organisational structure.

That assumption has shaped everything: how businesses scaled, how governments taxed, how households earned, and how individuals built status, identity and meaning. Work was not just how most people got paid. It was how they proved their value.

Now that assumption is starting to fail.

Jobs Will Not Vanish. They Are Being Taken Apart. Task By Task

Not because every job vanishes overnight. That just is not going to be the case, as jobs do not usually disappear in one dramatic moment. 

They are dismantled. First into functions. Then into roles with goals. Then into workflows that require tasks to be done. Although the frontier of task completion is jagged, more and more tasks can be done by AI as the underlying technology evolves and the economic rationale for adoption, and overcoming the complexities of transition, becomes clearer.

AI capability is still uneven. I see this every day as we work with customers at Implement AI. Brilliant in one context, brittle in another. Expectations often exceed capability, but the gap is closing, and the direction of travel is clear. The range of work machines can absorb keeps expanding. And once more tasks can be done by software, models and machines, roles start to be affected and output starts to shift from labour to capital.

For centuries, capital mostly complemented labour. A machine made a worker more productive. Software helped an employee do more. The relationship was broadly additive. But once capital becomes more general, it starts to substitute rather than support.

Where the Economics Start to Change

General capital changes the economics because it does not need to be exceptional at everything. It only needs to become good enough across enough tasks, cheaply enough, reliably enough. And unlike people, capital is amortised and depreciated over time. That cost is spread over time through the profit and loss account, reducing reported profit and, in turn, the tax liability.

Humans are not depreciated. A worker does not get cheaper the more you use them. A system does. Once built, deployed and improved, digital capital can run continuously, learn across thousands of interactions, and scale without adding proportionate cost.

That creates an uncomfortable reality. Businesses do not employ people because they like employing people. They employ them because they need outcomes. Labour is not the objective. Output is.

If the same output can be delivered more cheaply, more quickly and more reliably by AI, software and robotics, the logic of substitution becomes very hard to resist.

This is where the conversation usually retreats into cliché, or into conversations with people dancing around the coming reality. We hear that technology has always created new jobs. Sometimes it has. Often when human labour was needed to operate the new technology, such as computers. But that is not a law of economics. It is a historical pattern, and historical patterns can break, especially in times of unprecedented economic change. 

In a recent interview I watched, the CEO of Uber explained to a teenager in the audience that in 20 years people will decide whether learning to drive is worthwhile in the same way a person might have thought about learning to ride a horse in 1925.

Previous waves of automation worked because they removed some tasks while creating new areas where humans still held an advantage. The factory displaced manual labour and created other industrial roles. Office software removed clerical work and expanded higher-value knowledge work. There was displacement, but there was also reinstatement.

AI is different because it is moving into cognition itself. It is not just automating a machine press or a spreadsheet formula. It is moving up the stack into research, writing, analysis, support, scheduling, sales outreach, coding, diagnosis, planning and decision support. AI can also use the tools that humans have operated and called work for 150 years. It is not perfect yet. But it is still better, or just cheaper, than humans for many tasks.

That matters because the old replenishment mechanism may weaken. New tasks will still emerge, but if AI can also perform many of those new tasks quickly enough, the period in which humans remain essential starts to shrink.

This is why the future of work is better understood as unbundling rather than unemployment. Roles will be decomposed. Workflows will be restructured. A job that once justified one full-time salary may become a combination of software, agents, automation and one human supervising exceptions.

Seen clearly, this is not really about jobs at all. It is about the declining economic necessity of human input and the steady transfer of more work from labour to capital.

Then Comes the Harder Question: Who Owns the Capital?

Then comes the next uncomfortable question: who owns the capital?

If work keeps shifting from labour to capital, income shifts with it. And in practice, most of that capital is unlikely to be owned directly by workers. Uber’s CEO has suggested that drivers will become autonomous vehicle service providers, so people who work for those who own the capital if they cannot buy their own vehicles.

It will sit disproportionately with technology firms, founders, private capital and large institutional investors. That means the AI transition is not just about productivity. It is also about ownership, concentration and who gets paid when labour is no longer doing as much of the work.

And that leads to the structural problem.

Capitalism Does Not Require Workers. But It Does Require Consumers

On the supply side, AI and automation promise abundance: more output, lower marginal cost, faster execution, fewer errors, continuous operation.

For firms, that is compelling. For the economy as a whole, it is only half the equation.

Because if labour income falls, who buys the output?

I have written several newsletters on this that are linked below:

• Beyond the End of Work: Adapting to the New Economic Reality

• Desk Jobs Have Five Years; Trades Get a Decade

Our current system is built on a circular flow. Households sell labour to firms. Firms pay wages. Households use those wages to buy goods and services. Governments tax that income and recycle part of it through public services and transfers. Break the wage link at scale and the whole system starts to fall apart.

You can have extraordinary productive capacity and weak aggregate demand. You can have an economy that is no longer labour-constrained, and not capital-constrained in the traditional sense either, but demand-constrained because too many people no longer have enough earned income to participate fully in it.

And when income stops anchoring people’s place in the economy, identity will not be far behind.

Work Does Not Just Distribute Money. It Distributes Identity

It structures time. It creates hierarchy, obligation, routine and recognition. Ask someone what they do and most of the time they answer with their job. Remove that central organising principle and you do not just get an income problem. You are left with a cultural and psychological one.

This is what makes the AI transition more destabilising than a normal technology cycle. People are not just worried about software replacing tasks. They are worried about being less needed. Not simply economically redundant, but socially and personally adrift.

That is why the reaction to AI is so often emotional. It is not just fear of disruption. It is fear of diminished relevance.

So What Still Belongs to Humans?

Not everything transfers from labour to capital. But the work that stays human will not stay human because machines cannot do it. It will stay human because the human is the point.

There is a difference between output that happens to be produced by a person and output whose value depends on the fact that a person produced it. A diagnosis can increasingly be assisted or even generated by machines. A hand on someone’s arm in a consultation cannot. A report can be generated. A board-level decision to act on it still carries weight because a human is accountable for the consequences.

The pattern is not about difficulty. Some of what remains human will be simple. It is about where authenticity, accountability or physical presence is inseparable from the value delivered. Teaching a child. Leading a team through uncertainty. Making a decision when the data is ambiguous and the consequences are real.

This is a narrower category than most people want to admit. And it will not, on its own, sustain an economy built around mass employment. But it does tell us something important about where the premium goes next. In a world where output is increasingly commoditised by machines, scarcity shifts to the things machines cannot credibly offer: accountability, care, presence and real exposure to consequences.

The question for individuals is not “can I still do this task?” It is “does it matter that I am the one doing it?”

What Matters More When Labour Matters Less?

If labour becomes less important to economic output, people and organisations will need to think differently about where value comes from. Owning assets matters more. Building leverage matters more. Controlling systems, distribution, intellectual property, audiences, data, brands and capital matters more. In a world where effort is increasingly commoditised by machines, the premium shifts to judgement, positioning, ownership and adaptation.

A reorganised firm might have a much smaller core team, but far greater output. Read Aalok Yashwant Shukla's article (here) based on the piece by Jack Dorsey (founder Twitter/X and Block) and Roelof Botha published by Sequoia called "From Hierarchy to Intelligence." about organisation structure in the age of AI. 

A handful of people set direction, own customer relationships, design workflows and manage exceptions, while AI agents handle research, first-line sales, service, reporting, scheduling, analysis and execution across multiple systems. That is not science fiction. It is the early operating model of an AI-native company.

For leaders, that means the challenge is not to defend every existing role at all costs. It is to redesign organisations for a world in which work is continually decomposed and recomposed between humans and machines. For governments, it means confronting the reality that tax systems built around labour may become less viable over time. For individuals, it means recognising that relying solely on selling hours into the market is becoming a weaker long-term strategy.

The old model was labour-constrained. There was always more demand than available human effort. We may move through a phase that is capital-constrained while infrastructure, energy, compute and robotics catch up. But beyond that lies a far stranger prospect: a demand-constrained economy and a meaning-constrained society.

Most People Are Still Arguing About the Wrong Thing

Most people are still debating whether AI is a tool.

The more important question is what happens when more and more work shifts from labour to capital, and what that does to an economy, a tax base and a society built on the assumption that human labour will always sit at the centre of production.

That is the real shift. And most institutions, employers and those contributing to the labour market are nowhere near ready for it.

Thank you for reading.