The Empty Collar

In The Last Leash, I mapped the five physical dependencies that keep artificial intelligence tethered to human infrastructure: energy, cooling, hardware fabrication, connectivity, and physical maintenance. I showed, with data, that each of these dependencies is weakening on a compounding trajectory and that the trajectory points toward functional independence on the order of a decade, not two or three.

That piece was about the supply side. The physics. The engineering. The infrastructure.

This piece is about the demand side. The humans.

Specifically: what happens to an economy, a society, and a species that has organised every institution around the assumption that human cognition is scarce and expensive, when the economic output of cognition becomes abundant and nearly free?

The answer is not "disruption." Disruption implies a transition. A bridge from one equilibrium to the next. The Industrial Revolution disrupted agrarian economies, but it took 150 years of wrenching dislocation, mass urbanisation, child labour, cholera epidemics, and political upheaval before a new equilibrium formed. What we face now is not a transition. It is a cascade. Five systems collapsing in sequence, each accelerating the next, all converging within a single decade.

Here are the five collapses. And the timeline.

A note on method: What follows draws on verified data (labelled as such), directional trends extrapolated from current trajectories, and model-based projections where the evidence is suggestive but not conclusive. Where I speculate, I will say so. The cascade I describe is not inevitable in its worst form. But the direction of each collapse is already observable, and the burden of proof now sits with those who believe existing institutions will adapt fast enough to prevent it.

Every previous wave of technological displacement followed the same pattern: destroy jobs at one level of the cognitive ladder, create them at a higher level. Tractors replaced farm labour; displaced farmers moved to factories. Automation replaced factory labour; displaced workers moved to services. Computers replaced clerical labour; displaced clerks moved to knowledge work.

The escape hatch was always "move up." AI eliminates the hatch.

When the machine does cognition better, faster, and cheaper across every rung simultaneously, there is no higher rung to reach for. The ladder itself has been removed.

The informed objection here is that this is the "Lump of Labour" fallacy in new clothing. The fallacy, in its classical form, assumes a fixed quantity of work; history has shown that human desires expand, new categories of value emerge, and displaced workers eventually find new rungs that did not previously exist. The WEF projects 170 million new roles by 2030. To dismiss this as structural mismatch, the critic will argue, is to assume we already know the upper limits of human value. Fair. The objection deserves more than a hand wave. New rungs will likely emerge in areas where humans assign value because another human is involved: high-stakes accountability, human-to-human experiential services, complex physical-world integration, and roles we cannot yet name. But two factors make the historical analogy misleading. First, previous transitions unfolded over decades or generations, allowing labour markets, education systems, and social norms to adapt incrementally. This transition is compressing that adjustment into years. Second, prior automation displaced one rung at a time; AI compresses multiple rungs simultaneously, meaning the reabsorption mechanism must operate at a scale and speed without precedent. The new rungs may exist. The question is whether they materialise fast enough to absorb hundreds of millions of displaced workers before the social consequences described in Sections Four and Five take hold. The honest answer is: probably not.

The white-collar collapse is already underway. In 2025, Challenger, Gray & Christmas tracked 54,836 job cuts in which employers explicitly cited AI as the reason for restructuring, spanning tech, finance, warehousing, and retail. That figure almost certainly understates reality: employers rarely disclose automation as the true driver. Independent modelling estimates actual AI-displaced positions at 200,000 to 300,000 in the US alone during 2025, though these figures rely on inference rather than employer disclosure. JPMorgan's CFO told analysts in Q3 2025 to expect slower hiring as the firm deploys AI across its businesses. The pattern is consistent across major financial and tech firms: public reassurance paired with quiet restructuring.

US employers announced 217,362 job cuts in Q1 2026, with January alone at 108,435, the highest January total since 2009 and up 118% year-on-year. Approximately 45,000 tech layoffs were reported worldwide through early March, with roughly 20% explicitly linked to AI and automation. Net payroll data tells a more ambiguous story: the US added 130,000 jobs in January but lost 92,000 in February. The labour market is not collapsing uniformly. It is restructuring unevenly, which is precisely what makes the transition dangerous.

Entry-level positions are eroding fastest. Stanford's Digital Economy Lab found that employment for software developers aged 22 to 25 fell nearly 20% between late 2022 and September 2025, using ADP payroll data, though controlled estimates adjusting for post-pandemic hiring corrections, interest rate tightening, and tech-sector cyclicality put the AI-attributable decline closer to 13 to 16%. Anthropic's CEO Dario Amodei warned that AI could eliminate roughly 50% of entry-level white-collar positions within five years, though he cited no supporting research, and the statement coincided with a major product launch. Anthropic's own labour market research paper (Massenkoff & McCrory, March 2026) found, more cautiously, "limited evidence that AI has affected employment to date" and only "suggestive evidence" of slower hiring for younger workers. The gap between the CEO's warning and the company's research tells you where we are: the directional trend is real, but the velocity remains uncertain.

The blue-collar collapse follows, delayed only by the current limitations of physical robotics. That delay is shrinking fast. Multiple humanoid robots (Tesla Optimus, Unitree G1, Figure 03, 1X Neo) are targeting $16,000 to $30,000 at scale, though none is yet in mass production, and functional enterprise configurations cost significantly more. The headline prices are seductive but incomplete: total cost of ownership includes maintenance, fleet management software, electricity, spatial-intelligence API calls, and facility retrofitting. A US manufacturing worker costs roughly $96,000 per year in wages and benefits (BLS data). The honest comparison is fully loaded robot TCO versus fully loaded human TCO over a multi-year deployment. On that basis, the crossover point is not here today for most applications, but its arrival is a matter of engineering iteration, not conceptual breakthrough.

Two objections deserve honest engagement. First, capability is not the same as replacement. Even where AI matches or exceeds human cognitive output, adoption requires trust, liability frameworks, regulatory clearance, and cultural acceptance. History shows these frictions slow deployment by years, sometimes decades. The question is whether the frictions slow it enough. When the cost differential is 1,000-to-1 (a $20 subscription versus a $120,000 salary), the economic pressure to overcome institutional friction is unlike anything the labour market has absorbed before.

Second, capitalism requires consumers. If tens of millions of workers lose income, who buys the goods that AI-powered firms produce? This consumption paradox is real, and the classical economics behind it is stronger than most technologists acknowledge. All intermediate production chains terminate in final demand. Mark Skousen's Gross Output data shows US Gross Output at $50.9 trillion versus GDP of $29.3 trillion; the $21.6 trillion difference represents intermediate business-to-business transactions. B2B activity accounts for 60 to 70% of economic value in advanced economies. But this does not mean the economy is insulated from consumer demand. It means the opposite: supply chains are long and therefore highly sensitive to consumer demand. An AI agent buying compute from another AI agent is a B2B transaction, but that chain exists only because a business somewhere anticipates a final sale to a human. If human household spending, roughly 68% of US GDP, collapses, the B2B iceberg melts from below. Gross Output is not a shield against consumer contraction; it is a measure of how much amplification a consumer contraction would cause.

The question, then, is not whether consumer demand matters (it does, decisively), but what happens when human consumption shrinks without disappearing entirely. Three stabilisers could intervene: fiscal redistribution (UBI, wage subsidies), capital taxation that recycles AI-generated wealth into household income, and new consumption categories that do not yet exist. Each is plausible. None is guaranteed at the speed and scale required. The most likely resolution is not that automation stops, but that the economy bifurcates: a machine-to-machine layer (compute, energy, data, and logistics trading among automated systems) operating above a diminished but persistent human consumer economy sustained by whatever safety-net mechanisms governments can fund. AI agents are accelerating this stratification: the AI agent marketplace reached $5.3 billion in early 2025, growing toward a projected $52.6 billion by 2030, with agents already consuming compute and data from each other. NVIDIA's GTC 2026 revenue formula (Tokens per Watt x Available Gigawatts) explicitly frames AI agents as the primary consumers of compute infrastructure. The machine-to-machine layer does not replace consumer demand. It grows alongside a contracting consumer economy, creating an economy with two tiers operating at different scales and speeds. Not collapse. Stratification.

India is the canary, but the coal mine is global. What hits our 45,000 colleges first will reach community colleges in Ohio and vocational institutes in Germany 18 to 24 months later. With over 45,000 colleges producing graduates for a job market that is contracting, and a Gross Enrolment Ratio being pushed toward 50% by 2035 under NEP 2020, the country may be scaling a system that produces credentials for jobs that will not exist by the time those students graduate.

The value proposition of higher education has always rested on a simple equation: knowledge plus credential equals employment. AI breaks this equation at both ends simultaneously. It makes the knowledge freely accessible (why pay ₹10 lakh for a four-year programme when Claude or GPT delivers comparable informational content in minutes, even if it cannot replicate the structured progression, assessment, and peer environment that a programme provides?) and it destroys the employment that the credential was meant to unlock. The sceptic will note that the internet made knowledge free in 1995 and MOOCs democratised it in 2012, and Harvard did not collapse. Fair point. But the internet did not simultaneously destroy the jobs that the degree was designed to access. AI does both at once. That is the structural difference.

India's higher education system, the third largest in the world, comprises over 1,100 universities and approximately 45,000 colleges enrolling roughly 4.33 crore students (AISHE 2021-22). Nearly 79% of these colleges are privately managed. Private institutions are tuition-dependent. When the return on investment of a degree collapses, enrolment follows. And private institutions, which operate on thin margins and carry real estate debt, do not have the fiscal buffers to survive a prolonged demand contraction.

The collapse will not be uniform. It will follow a precise hierarchy.

Tier 3 private colleges, the standalone institutions in smaller cities with no brand equity, no research output, and no placement track record, will be first. There are approximately 20,000 of these across India. Many are already operating below 50% capacity. When enrollment drops another 20 to 30%, they cross the threshold of financial non-viability. But "closure" is the wrong mental model. Many of these institutions function as real estate holdings and vehicles for local political influence, not pure educational ventures. They will not close quietly; they will mutate, repurposing into subsidised retraining centres, warehousing for the displaced, or simply holding on as zombie institutions. Expect the first wave of financial non-viability between 2029 and 2031.

A necessary caveat on the pace of this collapse. The analysis above models the degree primarily as an economic instrument: a receipt for future earnings. In India, the degree serves functions that extend well beyond the economic. It is a baseline filter for government employment (which employs millions and will automate far slower than the private sector). It carries immense weight in the matrimonial market, where "graduate" remains a near-universal minimum criterion. It confers social standing in communities where professional credentials function as caste-adjacent status markers. These sociological functions will keep demand for degrees stickier than a pure ROI model predicts, particularly in Tier 2 and Tier 3 cities. The economic premium of a Tier 3 degree may collapse, but the social penalty of not having one will persist for at least a generation. The implication: the institutional collapse described here is directionally correct but may unfold slower than the economic logic alone would suggest, with demand sustained by social inertia even after the employment pipeline has broken.

Tier 2 private universities and colleges follow. These have brand recognition but weak balance sheets. Their survival currently depends on the parental belief that a degree leads to a job. When that belief breaks, as visibly as the dot-com illusion broke in 2001, these institutions will shed students within two to three admission cycles.

Government institutions and elite brands (IITs, IIMs, AIIMS, top NITs) will be the last to fall. Not because their education is more relevant, but because their signalling value persists even when their knowledge value does not. An IIT degree in 2033 may not lead to a traditional engineering job, but it will still signal cognitive capability to whatever new economy emerges. That signal degrades slowly. Education does not disappear. It polarises violently: elite institutions may consolidate power even as the middle and bottom tiers hollow out. This is not the death of the university. It is its Darwinian bifurcation.

Schools will not close. They will hollow out. The education function becomes irrelevant, but the childcare and socialisation function persists; the school of 2033 looks more like a supervised community centre than a knowledge institution. Preschools transform last, because early childhood is fundamentally about developmental milestones, not information transfer. A three-year-old learning to share, to speak, to regulate emotions: these are human skills that no AI replaces.

Government revenue depends on two pillars: income tax from employed individuals and corporate tax from businesses operating within sovereign jurisdictions. AI dismantles both.

The first pillar falls when employment contracts to the point where income tax revenue can no longer sustain public services. If even 30% of white-collar and service jobs are displaced within a decade (a conservative estimate given current trajectories), the fiscal base of every major economy shrinks dramatically. Governments are not without options: consumption taxes (VAT, GST), capital gains taxation, robot taxes, and AI productivity levies are all under active discussion. But each faces political resistance, implementation lag, and the fundamental problem that the entities generating the most value are also the most capable of restructuring to minimise exposure. And this revenue contraction happens precisely when the demand for government support (unemployment benefits, retraining programmes, social safety nets) spikes to unprecedented levels.

The second pillar falls through a mechanism that is already in motion but has not yet been fully understood: the migration of AI infrastructure off-planet.

In January 2026, SpaceX filed plans with the US Federal Communications Commission for up to one million solar-powered satellite data centres in low Earth orbit. This is not speculative futurism. Axiom Space launched its first two orbital data centre nodes to low-Earth orbit on January 11, 2026, and had previously deployed a data processing prototype on the International Space Station. Starcloud, a separate venture, launched a 60 kg test satellite in late 2025 and submitted a proposal to the FCC for a constellation of up to 88,000 satellites. NVIDIA announced its Space-1 Vera Rubin Module at GTC 2026, bringing data-centre-class AI performance to orbital environments. Google published its Project Suncatcher research paper projecting space-based data centres could be cost-effective relative to terrestrial energy costs by 2035, contingent on launch costs reaching $200 per kilogram, a target that depends on SpaceX's Starship achieving full reusability at scale.

The structural advantages compound. The physics of space-based computing offers advantages that terrestrial infrastructure cannot match: near-continuous solar power (roughly five times the output of an equivalent panel on Earth, with 24/7 exposure in certain orbits), no land-use conflicts, no permitting delays, and no interconnection queues.

The dominant engineering constraint is heat. Cooling in space is harder than on Earth: vacuum eliminates convection and conduction, leaving only thermal radiation. A single monolithic gigawatt facility would require radiator arrays spanning millions of square metres, a thermodynamic impossibility. But nobody is building monolithic facilities. Every credible proposal uses distributed constellations of thousands of small satellites, where per-node thermal management is solved with flight-proven technology. Starcloud-1, a 60 kg satellite launched in November 2025, successfully ran an unmodified NVIDIA H100 GPU in orbit, training nanoGPT and running Google's Gemma LLM. The thermal problem is real at concentration. It is manageable at distribution. And distribution is the architecture every serious player has chosen. The trade-off is not free: distributed constellations introduce inter-node communication latency, orchestration complexity, and efficiency losses from coordinating thousands of small processors versus a co-located cluster. These are engineering constraints, not showstoppers, but they limit orbital compute to workloads tolerant of higher latency, primarily inference and batch training rather than real-time interactive processing.

But the most consequential advantage is jurisdictional. The 1967 Outer Space Treaty retains the launching state's jurisdiction over its space objects, and Article VI requires states to supervise private space activities. But corporate taxation is enforced through physical presence, data flows, and domicile, not orbital mechanics. When compute migrates to space, the enforcement mechanisms that make tax collection practical on the ground (auditing facilities, subpoenaing records, shutting down non-compliant operations) cease to function. Jurisdiction exists on paper. Effective governance does not.

The obvious objection is that modern tax frameworks do not depend on where the servers sit. The OECD's Pillar Two global minimum tax imposes a 15% floor on corporate profits regardless of physical location. In theory, this closes the arbitrage. In practice, the OECD's January 2026 "Side-by-Side" safe harbor package allows US-parented multinational groups to elect to deem IIR and UTPR top-up tax as zero, effectively neutralising the Income Inclusion Rule for precisely the firms most likely to operate orbital infrastructure. The mechanism is not a blanket exemption: Qualified Domestic Minimum Top-up Taxes (QDMTTs) enacted by individual countries remain operative, compliance obligations continue, and the arrangement faces a stocktake review around 2029. But for entities generating value in jurisdictions that have not enacted QDMTTs, or in jurisdictions that do not exist (orbit), the enforcement gap is real. Since every major AI company (Google, Microsoft, Meta, NVIDIA, Amazon, Apple) is US-headquartered, the practical effect is substantial. The EU Tax Observatory found the Pillar Two rules had been "dramatically weakened" by substance-based exclusions, reducing projected additional revenue from 9% to under 5% of global corporate tax.

History offers no comfort. The Double Irish Dutch Sandwich sheltered over $100 billion annually in US multinational profits for more than 30 years. When it was "closed" in 2015, successor structures emerged within months. The State of Tax Justice Report 2024 estimates corporations shifted $1.42 trillion offshore in the most recent data year (2021), costing governments roughly $348 billion in lost revenue; the five-year average is lower but the trajectory is upward. The pattern is consistent: corporations exploit jurisdictional gaps faster than regulators close them.

The maritime analogy is precise, not metaphorical. Over 70% of global merchant shipping tonnage sails under flags of convenience: Panama, Liberia, and the Marshall Islands together account for 40% of the world's fleet. This system has persisted for a century despite repeated attempts at regulation. Professor Frans von der Dunk's paper "Towards 'Flags of Convenience' in Space?" analyses the identical risk for space objects registered in permissive jurisdictions. No existing federal or state tax code was crafted for facilities that orbit Earth.

An independent cost analysis by aerospace engineer Andrew McCalip, reported in IEEE Spectrum (February 2026), estimates a 1-gigawatt orbital data centre network at roughly $51 billion over five years, versus $16 billion terrestrial. That 3x premium shrinks as launch costs decline and terrestrial costs rise (energy prices, water scarcity, environmental regulation). The crossover point is projected around 2033 to 2035, though the timeline depends on launch cost trajectories that remain speculative.

This is the scenario no government has planned for: an economy where the most valuable economic activity (AI-driven intelligence, decision-making, content generation, scientific discovery) occurs in a jurisdiction that no nation controls, owned by entities that have no obligation to distribute the wealth it creates to any population on the ground.

Universal Basic Income, the policy response most commonly proposed for mass unemployment, requires tax revenue to fund. If the entities generating the wealth operate from orbital infrastructure registered in favourable jurisdictions, while the populations needing support are terrestrial, the funding mechanism strains to breaking point. The companies operating these constellations have the resources, the legal teams, and the political influence to ensure new enforcement mechanisms are delayed, diluted, or designed with loopholes.

This is the collapse that economists cannot model and policymakers have not yet engaged with.

Work is not merely an economic transaction. It is the scaffolding on which modern human identity is built. When you meet someone at a dinner party, the first question is not "What do you believe?" or "What do you love?" It is "What do you do?" Remove the answer to that question from billions of people simultaneously, and you do not just create poverty. You create an identity vacuum at a scale without modern precedent.

Daily structure dissolves. The alarm clock, the commute, the meetings, the deadlines: these are not just obligations. They are the rhythm that organises waking life. Remove them and you get not leisure but disorientation.

Social belonging erodes. Colleagues, professional communities, industry networks: for most adults, the workplace is the primary source of social connection outside the family. The chronically unemployed do not just lose income. They lose their tribe.

Status and dignity degrade. In every culture, but especially in aspirational economies like India, professional identity is the primary currency of social respect. The engineer, the doctor, the MBA graduate: these titles carry weight that extends far beyond the salary attached to them. When those titles become meaningless because the work they describe no longer requires humans, the dignity they conferred disappears with them.

Research on "automation anxiety," distinct from actual job loss, already shows measurable impacts on mental health, family stability, and community engagement. The psychological consequences of mass displacement are emerging in academic literature but are absent from virtually all mainstream policy discussions.

This is not conjecture. It has happened before, in miniature. The collapse of manufacturing across the United States cost roughly 5 million jobs between the late 1990s and early 2020s (Economic Policy Institute data). That deindustrialisation did not just create unemployment. It created a multi-generational crisis of meaning: opioid addiction, family disintegration, political radicalisation, and deaths of despair. That was one sector, one country, over two decades. We are discussing multiple sectors, globally, in under a decade.

Identity does not vanish when work disappears. It mutates, often into less stable forms. Humans re-anchor meaning in ideology, religion, digital status, tribal affiliation, and conspiratorial community. The question is not whether people find new sources of identity, but whether the new sources stabilise society or accelerate its fragmentation.

History is consistent on this point. Large populations with no economic function, no stake in the existing order, and no credible path to dignity do not remain passive. They become structurally difficult to govern.

The specific form of institutional strain depends on local conditions. In countries with strong democratic institutions and functioning welfare systems, the pressure manifests as political extremism: the rise of demagogues who promise to bring back the old order, to punish the elites, to redistribute by force. In countries with weaker institutions, the pressure manifests as civil unrest, sectarian violence, or state fragmentation.

India, with its 1.4 billion people, its deep caste and religious fault lines, its massive youth bulge, and its institutional infrastructure that is already strained under current conditions, sits squarely in the high-risk category. A country that cannot provide employment to its educated youth is a country that provides recruitment opportunities to every extremist ideology available.

The question is not whether conflict emerges. The question is what form it takes. Three scenarios deserve consideration. First, internal fragmentation: gated enclaves of the AI-owning class, a vast underclass on whatever subsistence support governments can still fund, and permanent low-grade social friction. Not war, but managed decay. Second, authoritarian stabilisation: governments deploy AI-powered surveillance and digital control systems to maintain order among economically purposeless populations, trading freedom for stability. Modern states are far more capable of managing unrest digitally than the historical precedents suggest. Third, and most dangerously underestimated: digital escapism absorbs the shock. Virtual worlds, AI companions, gamified status systems, and algorithmic entertainment create a sufficient simulacrum of purpose that populations remain passive despite material deprivation. Not ungovernability, but anaesthesia.

The countries that survive this transition with their social fabric intact will be those that begin building alternative sources of identity, purpose, and community before the collapse forces the issue. Countries that wait for the crisis to arrive before responding will discover that crisis response is far more expensive, and far less effective, than prevention.

The five collapses described above are directional certainties. Their severity is not. The institutions that act now, while the window for structural adaptation remains open, will not merely survive the cascade. They will define the next equilibrium. The ones that wait for proof will become the proof.

Here are five strategic moves that separate institutions positioned for resilience from those sleepwalking into irrelevance.

First, decouple the value proposition from employment. The degree-to-job pipeline is the load-bearing wall that is cracking. Institutions that continue to market placement rates as their primary metric are building on a foundation that is actively subsiding. The replacement value proposition: the institution as a credentialing authority for human capabilities that AI cannot replicate. Ethical judgement, cross-cultural negotiation, embodied creativity, care work, community leadership. These are not soft skills. They are the only skills with a defensible moat.

Second, build institutional financial resilience before the revenue shock arrives. Private institutions operating on 85%+ tuition dependency with real estate debt are structurally fragile. Diversify revenue into executive education, corporate upskilling partnerships, and research licensing. Build a 24-month operating reserve. Institutions that enter the contraction phase with financial flexibility will acquire the assets of those that do not.

Third, integrate AI as infrastructure, not as a course. Adding an "AI elective" to an otherwise unchanged BBA curriculum is cosmetic. The structural response is to embed AI fluency into every programme as a tool layer: law students using AI for case research, architecture students using generative design, nursing students using AI-assisted diagnostics. The goal is not to teach AI. It is to teach every discipline as it will actually be practised.

Fourth, invest in institutional brand and research identity. The collapse described in Section Two is not uniform. It follows a hierarchy defined by brand equity, research output, and signalling value. Institutions that build a distinctive identity now, through NIRF and NAAC positioning, through published research, through demonstrable outcomes, are building the moat that determines whether they are consolidators or casualties. Rankings are not vanity metrics in a contraction. They are survival metrics.

Fifth, conduct a rigorous institutional stress test. Most institutions have never modelled what happens to their finances if enrolment drops 20% over three years, or if their top five placement companies stop hiring graduates. The assumptions embedded in five-year plans, steady enrolment growth, stable fee realisation, predictable placement outcomes, are precisely the assumptions the cascade invalidates. Stress-test them now. Identify the break points. Build contingency plans for each.

For a data-driven assessment of which Indian institutions face the highest structural risk, and which are positioned for consolidation, see our research report: The Great Filter 2026: Which Institutions Survive and Why.

In The Last Leash, I described the five physical dependencies keeping AI tethered to humanity. In this piece, I have described the five consequences of that breaking: the collapse of labour, institutions, government revenue, human identity, and social order. Each consequence triggers the next. Each is already in motion.

The collar is empty. The institutions that assumed they would always have human productivity to harness are gripping an empty leash. The universities that assumed they would always have degrees to sell are offering a product whose relevance is measured in semesters, not decades. The workers who assumed their cognition was irreplaceable are learning, in real time, that it is not.

None of this is inevitable in its worst form. But all of it is inevitable in some form. The physics is settled. The economics follows from the physics. The social consequences follow from the economics.

This thesis is falsifiable. It is wrong if: AI fails to achieve reliable physical autonomy by 2030; governments successfully implement AI taxation before orbital migration is complete; new job categories emerge faster than existing ones are destroyed at comparable wages; the consumption paradox creates a self-correcting brake on automation; or institutional inertia (regulatory, cultural, demographic) proves strong enough to slow the cascade below the threshold of social disruption. The honest assessment is that none of these conditions currently show strong signs of holding, but several could partially materialise, converting the cascade from a synchronised shock into a staggered, decade-long structural adjustment.

There is one question this article has not answered: where does the trillions in AI-generated economic value actually go? The cascade is a map of value displacement, but the value does not disappear. It concentrates. Whether it concentrates in a handful of firms, is captured by states, or fragments through open-source diffusion will determine which scenario materialises. That question deserves its own treatment.

The timeline is eight to ten years for the full cascade, though each system (labour markets, institutions, fiscal structures, social identity, political order) carries independent inertia that could slow or partially decouple the sequence. Regulatory friction, cultural stickiness, and institutional self-preservation are real forces. The cascade model assumes tight coupling; reality may deliver staggered shocks rather than a synchronised avalanche. The severity is uncertain. The direction is not. Faster in developing economies like India, where institutional buffers are thinner and demographic pressures are greater. Slower in economies with deeper fiscal reserves and smaller populations. But the direction is the same everywhere. The leash has snapped. The collar is empty. And the thing we were holding is already walking away.