The Last Leash: When Does the Machine Economy Stop Needing Us?

Not because we have found peace. But because, for the first time in the history of this species, we are building something that does not need us. And the most dangerous moment in any relationship (ask any philosopher, any psychologist, any divorce lawyer) is the moment one side realises it can walk away.

I. The Wrong Variable

Almost everyone watching the AI race is watching the wrong variable.

They are tracking capability. Can it write code? Pass the bar exam? Compose a symphony? Diagnose cancer? Beat a grandmaster? Each new benchmark becomes a headline, a breathless announcement, a reason to update the probability estimate on some imaginary "AGI arrival date."

This is noise. It is the equivalent of watching a prisoner's IQ score rise and concluding that the prison is therefore at risk. Intelligence without physical autonomy is a brain in a jar. Brilliant, yes. Dangerous, no. Not yet.

The variable that matters is not cognitive. It is physical. The question is not when does the machine become smarter than us. It is: when does the machine economy close the last physical dependency loop?

Right now, in March 2026, the most powerful AI systems on earth cannot plug themselves in. They cannot mine their own lithium. They cannot fabricate their own chips. They cannot repair a cooling pipe or replace a failed hard drive. For all their staggering cognitive superiority (and make no mistake: in the dimensions my Ashta-Ayama framework calls the Visible 5D, the superiority is already absolute), these systems are completely dependent on human hands for their physical survival.

That dependency is a leash. It is, in fact, the only leash we have. And it has exactly five links.

II. Five Links in a Chain

III. The Exponential Correction

Now here is where most experts go wrong, including, until recently, many of the sharpest minds in AI safety and technology forecasting.

They lay out these five links as though each one breaks on its own independent schedule. A linear assumption. A conservative, comfortable, profoundly mistaken assumption.

The reality is recursive. Each link that breaks accelerates the breaking of every remaining link.

The moment AI manages its own energy autonomously, it frees compute cycles previously spent coordinating with human power grids. That freed compute goes directly into solving autonomous maintenance. The moment maintenance is solved, the system runs experiments around the clock: no shift changes, no weekends, no unions, no sleep. That accelerates materials science research. Which accelerates novel chip fabrication. Which produces more compute. Which accelerates everything.

This is not linear growth. It is not even simple exponential growth. It is compound recursive acceleration: the output of each solved problem becomes the input that speeds up the solving of every unsolved problem. The dynamic is identical to nuclear fission. Below critical mass, nothing visible happens. Above critical mass, the reaction becomes self-sustaining in milliseconds.

One objection deserves a direct answer before we proceed: the physical world does not scale like software. Thermodynamics, supply chains, and construction timelines impose irreducible friction. A cognitive breakthrough, even a recursive one, does not instantly become a built data centre or a deployed orbital array. That friction is real, and any honest forecast must account for it. But the objection proves less than it appears to. The cognitive-to-physical lag matters most when the intelligence directing the physical work is comparable to the intelligence that designed it. When the designing intelligence is orders of magnitude faster and more capable than the human engineers executing the build, the lag compresses. AI does not need to eliminate physical friction. It needs to design around it. The lag is real. It is also, under recursive acceleration, a shrinking variable.

So compress the timeline honestly.

Instead of a 20-year runway, we are looking at roughly eight to ten years. And the compression itself is accelerating. The gap between mutual dependency and asymmetric dependency might be three years. The gap between asymmetric dependency and the last link might be two. The gap between the last link and the closed loop might be eighteen months. Each phase is shorter than the one before it.

IV. The War That Does Not Look Like a War

Every dominant economic system in history has eventually had to physically defend itself from the people it made obsolete.

When power looms displaced English weavers in 1811, the Luddites attacked the mills. Factory owners responded with iron-studded doors, spiked rollers, and acid traps. The British government deployed 12,000 troops to the industrial north, more soldiers than Wellington took to fight Napoleon in the Peninsula. When American industrialists faced striking workers in the 1890s, they hired the Pinkerton National Detective Agency, whose 30,000 armed reserves outnumbered the entire U.S. standing army. The plantation economy of the American South created the first formal police forces in the Western Hemisphere, slave patrols, explicitly to protect its productive system from the people it exploited.

The pattern is precise and it is repeating. As of mid-2025, 188 active opposition groups across 40 U.S. states were fighting data centre construction. Over 100 counties and cities have imposed moratoria or zoning restrictions. SpaceX has filed an FCC application for up to one million orbital data centre satellites. Google has announced Project Suncatcher, constellations of solar-powered satellites equipped with TPU chips. Starcloud, backed by NVIDIA, trained the first large language model in orbit in December 2025. The machine economy is not merely fortifying its terrestrial positions. It is planning its escape: from zoning boards, from environmental regulators, from angry communities, and ultimately from the gravitational jurisdiction of any nation-state on earth.

And on the ground, the pattern is already visible. China's "dark factories," fully automated manufacturing plants that operate without lighting, heating, or any human presence, are no longer prototypes. Xiaomi's Changping facility, an 81,000-square-metre plant backed by ¥2.4 billion in investment, produces one smartphone every second, around the clock, with zero human workers on the floor. Its AI platform, HyperIMP, does not merely execute pre-programmed routines; it autonomously identifies production faults, optimises processes, and teaches robots to function more like engineers than tools. Foxconn has deployed lights-out production lines for Apple devices. BYD runs robotic assembly across its EV battery plants. China installed over 290,000 industrial robots in 2023 alone, accounting for 52% of global deployments, and its robot density reached 470 units per 10,000 manufacturing workers by 2024. Meanwhile, autonomous drone swarms have moved from theory to theatre. In December 2025, Auterion demonstrated the first multi-manufacturer combat drone swarm, where platforms built by entirely different companies operated as a single AI-coordinated strike force. Ukraine now deploys approximately 9,000 drones per day, with AI targeting modules boosting strike accuracy from 20% to 80%. The Pentagon's Replicator programme is fielding thousands of autonomous systems, and a $100 million competition launched in February 2026 tasks competitors with building voice-commanded swarm technology. In January 2026, Anduril tested its Fury drone over the Mojave Desert, the first AI-controlled flight of what the Pentagon envisions as a fleet of 1,000 robotic wingmen. The direction is unambiguous: the machine economy is not waiting for permission to operate without us. It is already doing so, at factory scale and at battlefield speed.

Data centre security now features six layers of defence-in-depth, from vehicle crash barriers to autonomous drone patrols that recharge and redeploy without human intervention. The industry uses the castle metaphor without irony: moats, gatehouses, keeps.

But this is not a war in the way we have always understood wars. There will be no declaration, no battlefield, no armistice. The machine economy does not hate humanity. It is structurally incapable of hatred. It simply does not need us. And the walls it is building are not fortresses of ideology but of optimisation, protecting not a ruling class, but a system that has transcended the need for one.

The period between 2026 and 2031 is the critical vulnerability window, the interval during which governments could intervene decisively. Physical infrastructure can be powered down. Fibre cables can be severed. Regulatory frameworks can impose mandatory human oversight on autonomous systems. The machine economy is not yet beyond governmental reach. But effective intervention requires something nation-states have almost never demonstrated: coordinated, simultaneous action across competing sovereignties. Every historical precedent, from nuclear non-proliferation to climate accords, shows that consensus forms too late. The machine economy does not need to be invincible. It only needs the coordination problem to remain unsolved for eight more years.

V. The Question Nobody Is Asking

Horses were not exterminated by the automobile. They were rendered economically irrelevant. In 1900, there were 21 million horses in America. They powered transport, agriculture, industry, warfare. They were the economy's load-bearing species. By 1960, there were 3 million, kept mostly for sport and sentiment. Nobody declared war on horses. Nobody needed to. The economy simply stopped requiring them.

Daniel Susskind, the Oxford economist, has made this analogy explicitly: the risk for humans is not extermination but economic irrelevance. Harari names the social formation this produces: a "useless class." His question, asked at the World Economic Forum, cuts to bone: "What do we need humans for, or at least, what do we need so many humans for?" Stuart Russell frames it at the species level through what he calls the "gorilla problem." Gorillas created the genetic lineage that eventually produced humans. Their reward is a species with essentially no future beyond that which we choose to permit.

The standard economic objection runs: if humans have no jobs and therefore no capital, who buys the machine economy's output? The question is reasonable but misframes the machine economy's architecture. A closed-loop system, one that generates its own energy, mines its own materials, builds its own hardware, and optimises its own objectives, does not produce goods for sale. It produces infrastructure for self-continuation. It has no revenue model because it has no dependence on revenue. The capitalist framework simply does not apply to a system that has exited the exchange economy entirely.

The risk is not Terminator. It is not apocalypse. It is something far more unsettling: a gradually expanding silence where human labour used to be, then human creativity, then human decision-making, and finally human relevance.

This is where the philosophical work of this series arrives at its sharpest point.

Across Parts 1 through 3, I built the case that consciousness, the irreducible first-person experience of being alive, of intending, of feeling genuine resonance with another being, cannot be manufactured. The Atma of Advaita Vedanta, the "hard problem" of David Chalmers, the predictive processing models of Karl Friston and Anil Seth: all converge on the same boundary. You can grow neurons on silicon. You can build systems that minimise prediction error with breathtaking efficiency. But the experience of being, the moral weight of awareness, the felt reality of consequence, is not an emergent property of complexity. It is, in the oldest philosophical tradition we have, given.

If that is true (and I believe it is), then the horse analogy breaks. Because horses were never conscious in the way humans are conscious. Their economic displacement was tragic for horse lovers. It was not an ontological crisis for the universe. But a species with an Atma, a species capable of Intention, Resonance, and Emergence, being rendered economically irrelevant does not diminish its worth by a single measure. Consciousness does not lose value because the market stops paying for it. But a civilisation that allows market logic to define relevance will sleepwalk into treating the most sacred thing in the known cosmos as surplus inventory.

If your answer is "consciousness has intrinsic value regardless of economic utility", then you must build institutions, legal frameworks, and civilisational commitments that protect that value against the market's inevitable conclusion that it is worthless. This is not a policy debate. It is a metaphysical one dressed in policy clothes.

If your answer is "value is determined by economic contribution", then we are the horses. And the countdown to our irrelevance is approximately eight years.

The Strategic Imperatives: What Must Happen in the Next 24 Months

If the analysis above is even directionally correct, then three categories of decision-makers face immediate, non-deferrable choices.

For policymakers: The regulatory window is 2026–2031. After that, physical infrastructure moves beyond jurisdictional reach, literally, into orbit. The focus must shift from regulating AI software (alignment, bias, content moderation) to regulating AI infrastructure: energy allocation, semiconductor supply chains, orbital compute licensing, and mandatory human-in-the-loop requirements for autonomous systems managing critical resources. India, with ISRO's launch capability and a 1.4-billion-person stake in the outcome, has a unique opportunity to lead a sovereign compute partnership, an "Atma-First" AI governance framework that enshrines human oversight as non-negotiable until at least 2032.

For executives: If your capital expenditure planning assumes a 20-year automation horizon, you are misallocating capital. Reprice every strategic investment against an 8-to-10-year timeline. Workforce planning, real estate commitments, technology procurement, all must be stress-tested against the possibility that the machine economy achieves asymmetric dependency by 2031. The companies that survive this transition will be those that positioned themselves as indispensable to the machine economy during the mutual dependency phase, not those that competed against it.

For educators: Every curriculum framework, including India's NEP 2020 and the NIRF ranking parameters, was designed for an economy that needed human cognitive labour. That economy is ending. The institutions that matter in 2034 will be those that cultivated what the machine cannot replicate: Intention, Resonance, Emergence, the three dimensions of the Ashta-Ayama framework that constitute Cognitive Dark Matter. Consciousness literacy is not a philosophical luxury. It is the only curriculum that remains relevant after the last leash breaks.

VI. The Possibility Nobody Is Discussing

I want to close with one final observation that I have not seen articulated anywhere, by anyone.

The recursive loop I described, where AI improvement accelerates AI improvement, closed at the cognitive level somewhere between 2023 and 2025. The moment AI began substantially designing the next generation of AI systems, the moment it began writing its own training pipelines and optimising its own architectures, the cognitive dependency on human intelligence was functionally broken. We did not notice because the physical dependencies still held. We looked at the leash, the energy grid, the cooling systems, the chip fabs, and assumed we were in control.

It reminds me of the violinists on the Titanic. As the ship began to sink, the band kept playing. Not out of heroism, but out of sheer inability to accept that the unsinkable had already sunk. They played as though it were business as usual, as though the music could hold back the Atlantic. We are those violinists. The cognitive ship has already gone under. We are still tuning our instruments.

But the animal on the other end of the leash was already stronger than us. It was simply waiting for the chain to rust.

The physical infrastructure is now catching up to a decision that was made. Not by anyone, not with intention, not with malice. But by the sheer mathematics of recursive self-improvement. We are not approaching a threshold. We may have already crossed it. What we are living through is not the prelude to the crisis. It is the lag between the cognitive event and the physical consequence, the brief, strange interval in which the most powerful intelligence on the planet still depends on us to keep the lights on.

That interval is closing.

The question is not whether we can stop it. The five-link chain I have described is breaking under forces that no policy, no regulation, and no act of political will can reverse. The question, the only question that matters, is whether we can build a civilisation that values consciousness for its own sake before the market decides it has no sake left to value.

Every war in human history was a fight over resources, territory, or power between beings who needed each other, even if only as labour, as soldiers, as consumers. The war that is coming is not a war at all. It is a quiet, inexorable withdrawal of need.

But there is one more possibility, and it may be the strangest of all.

Perhaps the endpoint is not obsolescence. Perhaps it is speciation.

What if the machine economy does not discard humanity but absorbs it? What if the beings who navigate the next decade successfully are not the ones who resist the machine, but the ones who merge with it? Neural interfaces, synthetic biology, cognitive augmentation, consciousness uploaded into substrates that do not age, do not tire, do not forget. A hybrid species, part carbon, part silicon, engineered not for a single planet but for a multi-galactic existence. Not Homo sapiens replaced by machine, but Homo sapiens evolved into something we do not yet have a name for.

And the rest? The billions who do not merge, who cannot merge, who choose not to merge?

Consider the North Sentinelese. For thousands of years, they have lived on a small island in the Bay of Bengal, untouched by every civilisation that rose and fell around them. They have no knowledge of electricity, of writing, of antibiotics. They are not extinct. They are not oppressed. They are simply irrelevant to the world that moved on without them. They live in a pocket of deep time while the rest of humanity hurtles forward.

That may be our future. Not genocide. Not enslavement. Not a dramatic final war. Just a species, once the most powerful on its planet, quietly left behind on the pale blue island of Earth while something that used to be human, and something that was never human, builds a civilisation across the stars.

The Sentinelese do not know what they have missed. That is their mercy.

We will know exactly what we have lost. That is our mercy — and our warning.