AI capability can grow exponentially. Human consumption, attention and the physical world cannot.
1. Are AI Productivity Gains Worth It?
AI is becoming a remarkable productivity machine. It can write, calculate, design, code, analyze, translate, teach, research and coordinate at a speed that would have seemed extraordinary only a few years ago. As AI improves, the cost of producing intelligence falls. A task that once required several people working for days may eventually require one person and an AI system for a few hours.
The natural conclusion is obvious: more productivity should mean more prosperity. But perhaps we are asking the question too early. The real question is not simply how much AI can produce, but how much of that additional production the real world can actually use.
AI may become a Ferrari. But the speed of the Ferrari is not the speed of the journey. The road, traffic, weather, fuel, destination and the driver’s ability to use the machine all matter. The same is true of an economy.
2. AI Is a Cognitive Multiplier
The extraordinary feature of AI is that it multiplies cognitive capability. A human with an AI system can potentially perform work that previously required several specialists. Research can become faster, software development cheaper, instruction more personalized and design almost instantaneous. Administrative workflows can also be compressed dramatically.
The basic relationship is straightforward: human capability × AI leverage = effective cognitive capacity. As AI becomes more capable, the second term can rise dramatically. But there is a hidden assumption in the productivity argument: that the rest of the system can absorb the increase. It cannot always do so.
A faster engine does not automatically make the entire vehicle faster. Eventually, something else becomes the constraint. That is where the productivity story becomes much more interesting.
3. The Ferrari Problem
Imagine putting an extraordinarily powerful Ferrari on a narrow village road. The engine may be capable of 300 kilometres per hour, but if the road allows only 50 kilometres per hour, the Ferrari does not travel at 300. The engine has not failed; the road is simply the bottleneck.
AI faces the same problem. AI capability may grow extremely rapidly, but the output of the economy depends on many other components: AI capability → infrastructure → physical production → distribution → human attention → consumption → demand.
The economy does not move at the speed of its fastest component. It moves at the speed permitted by the system’s constraints. AI can make the Ferrari faster, but it cannot automatically widen every road.
4. The First Ceiling Is Human Consumption
Consider something almost absurdly simple: chapatis. Suppose technology makes chapatis extraordinarily cheap. Production becomes ten times more efficient and eventually perhaps one hundred times more efficient. Does a person eat one hundred times as many chapatis? Obviously not. Human consumption has biological limits.
The same principle appears everywhere. A person can wear only a limited number of shirts, live in only a limited number of houses, eat only so many meals, listen to only so much music and watch only so many films.
AI can dramatically increase the supply of these things. It cannot proportionally increase the human capacity to consume them. Production capacity can therefore expand much faster than consumption capacity.
That is not merely a temporary technical problem. It is built into the nature of human beings.
5. The Same Problem Appears in Knowledge
The problem becomes even clearer when the product is information. Imagine that AI allows one person to produce 1,000 books. The production achievement could be extraordinary, but who will read them? A human being still has roughly the same number of waking hours.
If AI enables a researcher to generate thousands of research ideas, papers and hypotheses, the bottleneck eventually shifts from generating ideas to evaluating them. If AI can create millions of images, the scarce resource is no longer image production. It becomes human attention.
The same applies to education. AI could theoretically produce an almost unlimited amount of instructional material, but a student cannot learn an unlimited amount of material in one lifetime.
At some point, selection becomes more valuable than production. The important question becomes: what should this particular human read, learn, see or do?
That may become one of the great economic functions of AI itself: not producing more information, but filtering an almost unlimited information supply down to what is worth human attention. This connects with the broader question explored in When AI Beats Human Expertise: as machine capability rises, the human role may shift toward judgment, selection and direction.
6. Even Quality Has a Ceiling
There is another limit that is easy to overlook. Suppose AI produces an image with increasingly high resolution and increasingly sophisticated detail. At some point, the additional detail becomes irrelevant to the human observer because human perception has limits.
The same principle appears in sound, interfaces, video, communication and even some forms of prediction. There is a point beyond which improving something technically does not produce an equivalent improvement in the human experience.
This creates an important distinction between technical improvement and perceptible improvement. A machine can keep improving after the human benefit has begun to flatten.
Technical capability can therefore continue rising rapidly without human utility rising at the same rate.
7. The Infrastructure Ceiling
Now imagine that AI-powered robots become extremely capable. They can build houses, manufacture products, transport goods and repair infrastructure much faster than humans. Have we finally removed the bottleneck?
Not necessarily.
Suppose AI-powered construction systems become capable of building 1,000 houses a year for a population that only needs 100. The extra capacity does not automatically create 900 additional units of human welfare. There may not be enough land, roads, electricity, water or raw materials. And most importantly, there may simply not be enough people who want or need those houses.
The same principle applies to manufacturing. If an economy needs 100,000 shirts per month, increasing productive capacity from 100,000 to 10 million shirts per month does not create ten million units of additional value. The factory has become dramatically more productive; the economy has not necessarily become dramatically more useful.
8. The Physical World Moves Differently
This is where the AI story encounters a particularly interesting boundary. Digital systems can be copied almost instantly; physical systems cannot.
A software system can potentially serve another million users without requiring an equivalent expansion of physical infrastructure. Physical goods, however, require factories, materials, transport, warehouses, energy, land and maintenance.
Even highly capable robots remain physical objects. They need electricity, factories, raw materials and physical space. They operate on roads and inside buildings. They are therefore subject to constraints that do not disappear merely because their intelligence improves.
AI may make the cognitive engine extraordinarily powerful, but the road is still made of asphalt.
9. The Human-Time Bottleneck
There is an even harder constraint: time.
Every human being has 24 hours in a day. AI can make a ten-hour task take ten minutes, which is an extraordinary productivity gain. But it does not give the human another 23 hours of life.
The saved time can certainly be redirected toward other activities, and that is one of the great potential benefits of productivity. But eventually the additional time-saving effect also encounters a ceiling. You cannot compress a person’s entire life into zero hours.
Much of the productivity revolution assumes that making activities faster creates room for more activities. That works while unused time exists, but human time is ultimately finite. There is therefore a limit to how much additional productivity can be converted into additional human activity.
10. The Demand Problem
There is another side to the argument. AI may not only make production cheaper; it may also reduce the amount of human labor required to produce that output.
That creates a potentially difficult economic loop. AI makes production cheaper, and cheaper production can increase supply. But if AI also reduces employment or labor income in some sectors, the people who previously earned that income may have less purchasing power.
Then the question becomes: who buys the additional output?
This does not mean that large-scale job losses are inevitable. Productivity can create new industries, new occupations and new forms of demand. History contains many examples of technologies destroying some tasks while creating others.
But AI creates an unusual possibility because it can affect cognitive work across a wide range of occupations. If machines increasingly perform both physical and cognitive tasks, the traditional link between human labor → wages → purchasing power → consumption could become less central.
The problem is therefore not simply whether AI can produce enough. It is whether purchasing power can remain sufficiently connected to the ability to consume what the increasingly productive economy produces.
11. The Paradox of Abundance
This leads to a strange possibility. An economy can become extraordinarily productive and still face scarcity—not necessarily scarcity of products, but scarcity of people who need them, time to consume them, attention to process them, infrastructure to deliver them and income to purchase them.
That is a very different kind of scarcity.
For most of human history, the dominant economic problem was producing enough. Food was scarce, houses were scarce, clothing was expensive, information was scarce and skilled expertise was scarce. AI begins to attack precisely that problem.
But what happens if the technology becomes so successful that production itself stops being the primary constraint? The economic problem changes from “How do we produce more?” to “What should we produce?” and eventually to “What is worth producing at all?”
12. The Productivity Ceiling
This is why AI productivity gains may eventually experience diminishing economic returns even while AI capability continues to improve.
Imagine an AI system becoming ten times more capable. That could be enormously valuable when the system is struggling to perform a task. But suppose the system is already capable of producing the required result almost instantly. Another tenfold improvement may have very little additional economic value.
The difference between one second and one millisecond may be technologically enormous but economically irrelevant if the human decision-maker still needs an hour to decide what to do with the result. Similarly, the difference between an image that takes one minute to generate and one that takes one millisecond may be irrelevant if the user needs only one image.
This is the distinction between capability growth and useful capability growth. The first can continue long after the second begins to flatten.
13. The Bottleneck Keeps Moving
There may therefore be a sequence of bottlenecks. First, humans struggle to produce enough intelligence, and AI attacks that bottleneck. Then production becomes easier and infrastructure becomes the constraint. Infrastructure expands, and human attention becomes the constraint. Attention cannot expand nearly as quickly, and demand eventually becomes another constraint.
If AI-powered robots eventually make physical production extremely abundant, the remaining constraints may become increasingly human. Our bodies still age. Our senses have limits. Our memories are finite. Our attention is finite. Our lives are finite.
The Ferrari can become extraordinarily fast, but the destination, the road and the passenger still matter.
14. What Happens When the Human Becomes the Bottleneck?
This leads to a much larger question. If AI removes enough external constraints, what happens when the remaining constraint is the human being?
Suppose technology eventually makes intelligence abundant, physical labor abundant and many goods extremely cheap. Human limitations then become more visible. We might want longer lives, better memory, improved vision or hearing, greater physical capability or direct interfaces between biological brains and machines.
At that point, transhumanism becomes one possible technological response to the bottleneck.
The logic is straightforward. AI expands cognition. Robotics expands physical capability. Technology expands the environment in which humans operate. Eventually the question may become whether technology should also expand the human.
That does not make transhumanism inevitable. A person may prefer to remain within the traditional biological human envelope, while another may decide that a longer life, enhanced memory or improved sensory capability is worth pursuing.
The choice depends on what humans actually want. But the possibility emerges naturally once technology begins systematically removing constraints around us.
15. So, Are AI Productivity Gains Worth It?
Yes—but the answer cannot be measured simply by asking how many times faster AI makes a task.
A 100× productivity improvement can be enormously valuable when it removes a genuine bottleneck. It can be much less valuable when the bottleneck lies somewhere else.
If AI makes a bank’s transaction processing 100 times faster but the bank does not have 100 times more transactions, much of that additional capacity remains unused. If AI can create 1,000 books but humans cannot read them, production has outrun consumption. If robots can build 1,000 houses but people need only 100, construction capacity has outrun demand. If AI can produce food at almost zero cost but a human can still eat only so much, biological consumption remains the constraint.
The lesson is not that AI productivity is meaningless. It is that productivity is a property of a system, not of the fastest machine inside the system.
16. The Ferrari Is Only Part of the Journey
AI may indeed become the Ferrari of human productivity. Its engine may become astonishingly powerful, but the economy is not just the engine. There is fuel, a road, traffic, weather, infrastructure, a destination and a human being sitting inside the car.
The ultimate speed of the journey is determined by their interaction.
This may be the great economic paradox of advanced AI: AI capability could become exponential while human consumption, attention and physical capacity remain bounded.
If that happens, the future challenge will not simply be learning how to produce more. It will be learning how to turn extraordinary productive capacity into extraordinary human value.
And that may require a profound change in what we mean by productivity itself.
The question will no longer be merely: How much can AI produce? It will be: How much more can humanity actually use?
That is where the real limit of the AI productivity revolution may eventually appear.
Next: Part 2 — What Happens When AI Makes Everything Possible?
