Where will the money come from?
AI can already perform 80–90 per cent of the tasks involved in much of my own commercialization work.
Not perfectly. Not autonomously in every situation. And not without me remaining responsible for objectives, judgment, relationships and consequences.
But the practical reality is becoming difficult to ignore.
Research, analysis, planning, writing, segmentation, account preparation, campaign development, reporting and much of the operational follow-up can already be performed by AI.
We have examined legal work and found much the same pattern. The same question applies to accounting, auditing, administration, case handling, consulting, marketing and a growing number of other knowledge-intensive occupations.
If AI can perform 80–90 per cent of these tasks today, what happens as more defined workflows move closer to 100 per cent?
The first consequence is obvious
The cost of producing knowledge work will fall dramatically.
When something that previously required a team, several weeks and a substantial budget can be produced by one person working with AI in a matter of hours, buyers will eventually stop paying as if the old production model still existed.
They may still pay for accountability.
They may pay for exceptional judgment, trusted relationships, access, reputation, risk-taking and the willingness to stand behind a decision.
But they will not continue paying the same price for hours spent gathering, processing, formatting and communicating information when machines can do most of that work almost instantly.
This does not merely threaten individual jobs.
It threatens the economic logic of entire professions.
The second consequence is more important
Where will the money come from?
Most of the discussion about AI concerns productivity: how much more can each person or company produce?
Far less attention is given to demand: who will have the income required to buy everything we become capable of producing?
A large part of the Western economy is built around people producing, interpreting, distributing, controlling and applying knowledge.
The IMF estimates that around 60 per cent of jobs in advanced economies are exposed to AI. Across OECD countries, roughly three out of four workers are employed in service industries. The ILO finds that clerical occupations remain the most exposed to generative AI, while exposure is also increasing in highly digitized professional and technical work.
Exposure does not mean that all these jobs disappear.
Most occupations contain a mixture of automatable tasks, human responsibilities and physical or relational work.
But entire jobs do not need to disappear for the economic effect to become enormous.
If one person can do what previously required five, companies do not need to remove every employee. They only need to stop hiring four of them.
If professional services become substantially cheaper, revenue, wages and employment can fall even while output and productivity rise.
And the consequences will not remain inside law firms, accounting practices, consultancies or corporate offices.
The people working there pay builders, taxi drivers, restaurants, hotels, airlines, retailers, tradespeople and countless other businesses.
When their income disappears — or simply becomes less secure — they spend less.
A restaurant cannot pay its employees with productivity statistics.
A construction company cannot build houses for customers who no longer qualify for mortgages.
An airline cannot fill its seats with demand that exists only in theory.
If people do not have money, demand falls.
Productivity does not automatically become prosperity
The optimistic answer is that technology has always destroyed some jobs and created others.
That is true.
But it does not answer three questions:
1. Will new work emerge as quickly as existing work is reduced? 2. Will it employ roughly the same number of people? 3. Will it distribute purchasing power widely enough to sustain demand?
AI differs from many earlier technologies because it is not aimed at one profession, industry or type of physical labour.
It increasingly operates across the cognitive tasks found inside almost every modern organisation.
It can also improve through better models, tools, data, memory and orchestration without requiring each company to rebuild the underlying technology.
The transition may therefore be faster and broader than our institutions, labour markets and education systems are prepared for.
Perhaps entirely new industries will emerge.
Perhaps falling prices will increase consumption.
Perhaps human expectations will expand and create work we cannot yet imagine.
I hope so.
But hope is not an economic mechanism.
The real question is distribution
AI may create extraordinary wealth.
The problem arises if that wealth is concentrated among the relatively small number of people who own the models, infrastructure, data, capital and highly scalable companies while the income of everyone else declines.
An economy does not function merely because value is created.
It functions because purchasing power circulates.
If AI allows ten people to produce what previously required one hundred, we must eventually decide how the productivity gain reaches the other ninety — not only as cheaper products, but as actual income, ownership or economic participation.
That could involve shorter working weeks, broader ownership, profit-sharing, new forms of taxation, universal services, income guarantees or models we have not developed yet.
I do not know which answer is right.
I am not even certain that we fully understand the question.
But I am increasingly convinced that “AI will make us more productive” is only the beginning of the discussion.
The more urgent question is this:
When machines can produce most of what people currently receive wages for producing, where will the money required to buy that production come from?
What changed
I used to think primarily about AI as a production question.
How much more can one person do? How quickly can a company move? How much professional work can be performed at a lower cost?
Those are still important questions.
But the more of my own work AI can perform, the more the problem moves from production to demand.
The economic risk does not begin when AI replaces every task or every job.
It begins when productivity rises fast enough that fewer people are needed to create the same value, while the gains are not distributed widely enough to replace the income and demand that disappear.
What I think matters
We need to stop treating productivity as if it automatically produces shared prosperity.
Companies should examine honestly how much work AI can already perform. Individuals should do the same. Avoiding that analysis will not protect anyone from competitors who complete it.
But governments, employers, investors and the companies building AI must also examine the other side of the equation.
If labour is no longer the main way most people receive a share of economic production, what takes its place?
That is not a distant philosophical question.
It is a commercial question about who will be able to afford the products and services companies expect to sell.
What remains uncertain
AI exposure is not the same as job replacement. Many occupations will change rather than disappear. New jobs, industries and forms of demand may emerge. Falling costs may make previously expensive services available to far more people.
We do not yet know how quickly companies will reorganise around the capabilities that already exist, how much human involvement customers will continue to value, or how strongly regulation and institutional inertia will slow the transition.
We also do not know whether the productivity gains will be distributed through wages, ownership, lower prices, public services or entirely new economic arrangements.
What seems increasingly difficult to defend is the assumption that the old distribution of income can remain unchanged while the need for human knowledge work falls dramatically.
AI may solve the problem of producing more.
It does not, by itself, solve the problem of who can pay for it.