The decade ahead is a dystopian one, irrespective of whether you sit on the techno-optimist or the doomsayer side of the debate. The fundamental relationships that hold up our economy are changing right
before our eyes, and the system has not even begun to acknowledge it. These cataclysmic shifts are unfolding in real time, and because they are unfolding in real time, we cannot see their full impact on our society, economy and lives.
Take the relationship between capital investment and job creation. This relationship has been broken for over a decade. Every other country has at least recognised it; some have started to address it. In India, the system continues to behave as if the 1990s never ended. States hand out land, incentives and tax holidays based on the size of the capital investment announced, not the number of jobs actually created. Every year, each state hosts its own investment festival, where top industrialists are almost forced to announce large investments. A few years ago, while running a multi-state newspaper, we did the math: the actual investment was less than 1 percent of the announced number in the first year, and less than 0.1 percent by the second. That is a different scandal.
Coming back to the broken relationship between capital invested and jobs created — if the investment is in manufacturing, the higher the capital deployed, the more automation on the shop floor, and the fewer people employed. A semiconductor plant is the textbook illustration. Because of the strict air-conditioning parameters — no pollutants are allowed — there is nobody on the shop floor. The handful of people employed are all behind consoles. Tens of billions of dollars in capital equipment result in direct employment of fewer than 100 people at the plant level. The same trend is visible from cement plants to steel plants. Automation is rising rapidly, human headcount is dropping exponentially, and the combination of AI and robotics is going to push this even further. This is not a future scenario. It is the current reality. Automation in manufacturing is already in its fourth generation. AI in services is in its third. Both are now growing exponentially.
This brings us to the dishonest comparison that AI optimists keep deploying — that we have seen this before, that every wave of technology has destroyed some jobs and created more. The shift from agriculture to manufacturing took almost 170 years. There was time for multiple generations to prepare, retrain and adjust to this shift. A farmer’s grandson could become a factory worker. A factory worker’s grandson could become a software engineer. The transitions were generational, and the entire social architecture — schools, universities, labour laws, welfare states — was built around the assumption that change happens at human speed.
When people invoke that history to dismiss the present, they forget three things.
One, the steam engine was not an exponential technology in the way AI is. It did not iterate and improve itself with every exchange with humans on an exponential basis. It did not get cheaper, faster and smarter every six months because of its own use. AI does. Every conversation, every query and every deployed model become training data for the next, more capable model. There is no comparable feedback loop in the history of industrial technology.
Two, every previous technology gave humans time to adapt, learn and adopt it, so that they were not left out of the workstream or the earning stream. When the assembly line arrived, the worker did not vanish overnight; he was redeployed, retrained, absorbed. AI does not give this runway. The model that replaced the call-centre agent in 2023 has already replaced the L1 coder in 2025 and is gunning for the L3 analyst.
Three, the earning stream from previous technologies was not consolidated with a few companies that captured all the upside benefits of adoption while throwing humans out of the loop. Steam and electricity diffused. Railways and telecoms were eventually regulated as utilities. The upside got shared, however imperfectly, across capital, labour and the state. AI is being built and owned by a handful of companies that capture nearly all the upside while pushing humans out of the loop. The diffusion model that made previous transitions tolerable does not apply here.
A similar trend is already playing out in the IT and BPO sectors. The rise in digital crime — digital arrests, email and WhatsApp fraud, fake call centres — is, in my reading, a direct consequence of job disruption in the BPO sector. Most organisations, CXOs and industry associations are not prepared for the societal-level disruption and crime wave their automation has unleashed. BPO was the first sector to adopt RPA, chatbots and AI in its processes. This led to the firing of a large number of customer service employees trained, very specifically, to sell services over the phone and extract information from strangers. Most BPOs today are a fraction of their pre-COVID employment peak. That trained workforce did not vanish. It went somewhere. A significant slice of it has turned to crime because it could not find honest work that paid even half as well.
The displacement effects of automation are not contested in serious economic literature. Daron Acemoglu — who has since won a Nobel — laid them out clearly in a 2019 paper that has aged disturbingly well. He noted, in essence: the presumption that all technologies raise aggregate labour demand simply because they raise productivity is wrong. Some automation technologies entail sizable displacement effects with only modest productivity gains, particularly when the displaced workers were cheap to begin with and the automated technology is only marginally better than them. Because of this displacement effect, we should not expect automation to create wage increases commensurate with productivity growth. Automation by itself always reduces the labour share of industry value added and tends to reduce the overall labour share in the economy. The reason we historically saw stable labour shares is that other technological changes simultaneously generated new labour-intensive tasks. Whether labour retains a role in production depends on whether that counterbalancing process continues — and there is no guarantee that it will.
The point Acemoglu makes carefully, with footnotes and equations, is the point our policymakers, economists and CXOs continue to ignore. In manufacturing, the use of capital to substitute labour has been studied since the 1980s. When you point out that AI is doing the same substitution in services, you are told it cannot happen. The same people who watched welding robots replace welders insist that no model can replace a coder, an analyst, a paralegal or a radiologist. They cannot see that the trend that replaced labour with machines on the shop floor is now replacing labour with software at the desk.
Meanwhile, every person with a keyboard is producing uninformed opinions on LinkedIn and in WhatsApp groups. Your opinion is yours; it is not a fact, and it is not true simply because your under-researched brain finds the past data inconvenient and refuses to engage with it. There is now a new breed of commentator — the ones who make up their opinion first and then go to an AI to write their post for them. These AI-opinion uncles have forgotten that writing is thinking only if you do the writing yourself. And even when you do, if your brain does not have the data, the output is still muddled.
Muddled thinking is not limited to AI uncles, though. It runs through the entire meritocratic class — the same class that runs our systems but cannot project a trend it has not been spoon-fed by a professor or a teacher. The fact that we are entering a dystopian decade of exponential labour displacement is a trend this class cannot see, because a dystopian outcome is hard to imagine when your career has been built on a steady diet of optimism. Anything that puts a dent in the optimism bubble is filed away as negativity and ignored. I know this because I have been having this exact conversation with policymakers, bureaucrats, industry leaders, CXOs, technologists and association heads every week, every day. Everyone wants to hide behind cheerful historical analogies their chattering brains can conjure up. Nobody wants to sit with the discomfort that exponential technology has been reshaping our lives, society and economy at a pace our linear brains have not caught up with.
The misunderstanding most aggressively promoted — including by AI companies themselves — is that there will be net job creation. The tech-bro version of this argument shows up in different costumes. The latest one is that replacing COBOL with AI is a trillion-dollar opportunity, since AI has brought the cost of a line of code down from a thousand dollars to two or three. What this argument conveniently hides is the nature of the technology itself. AI is exponential. If it can crash the cost from a thousand dollars to two dollars, it can crash it from two dollars to two cents. And if AI is doing the work, it will replace coders faster than firms can retrain them.
But the conversation cannot end at jobs, because jobs are only the first domino.
When you push humans out of the manufacturing line, they migrate to construction and gig work. When you push BPO workers out and they cannot find anything else, some of them turn to crime. Now ask the next question, the one the tech bros do not want asked: what happens when humans are thrown out of the earning loop entirely? What happens when the economic system itself becomes substantially independent of human labour?
The first thing that breaks is not the labour market. It is the state.
A state runs on tax collected from human income — directly through income tax, indirectly through consumption taxes that depend on human earning power. AI companies, almost without exception, sit outside this taxation structure. They are headquartered in low-tax jurisdictions, they book profits through transfer pricing across continents, and they have optimised their structures so thoroughly that no single state has meaningful jurisdiction over them. They have engineered themselves to be largely uncontrollable by any single national government.
Now stack this against an AI-driven displacement of human labour. Humans earn less or stop earning. They pay less tax, or no tax. They become dependent on the state for welfare and a basic income. The state, in turn, earns less from its traditional tax base and becomes increasingly dependent on these supra-state AI companies — for revenue, infrastructure and even the basic plumbing of public services. The balance of power inside a democracy then shifts from the elected state to a handful of AI companies that are nominally based somewhere but answerable nowhere. The democratic system, which assumed that the state was the most powerful actor in the room, quietly stops working in the way the textbooks said it would.
This is not a doomsday scenario. It sounds extreme only because we have not named it. Some variation of this dynamic is already underway in the United States, Taiwan and South Korea, where a small number of technology and semiconductor companies have grown larger and more strategically critical than entire ministries. The state in each of these countries is increasingly negotiating with, rather than regulating, its own corporate sector. The direction of dependency has quietly reversed.
The workers who once found jobs in manufacturing and construction have already become gig workers. The BPO workers who lost their jobs and could not find new ones have, in too many cases, turned to crime. Will the displaced coders also turn to white-collar and AI-enabled fraud as their jobs disappear? Will the state, slowly starved of revenue, find itself unable to police them even if it wants to?
Are we going to see a rise in crime, and a hollowing out of the state itself, in the dystopian decade ahead?
