On Productivity, the Internet's Unfinished Lesson, and Why AI May Be the Exception That Breaks the Pattern
"It is the case that productivity creates more jobs. Just go back and look at history." — Jensen Huang, CEO of NVIDIA
I. The Mundane Oracle
There is a particular kind of genius that chooses the most ordinary object available to explain the most extraordinary moment in human history. Jensen Huang, the founder and chief executive of NVIDIA — the company that built the computational infrastructure upon which the current artificial intelligence revolution runs — chose a washing machine.
Not a steam engine. Not a microchip. Not a server farm. A washing machine.
He could have invoked the grandeur of the Industrial Revolution or the sublime complexity of neural networks. Instead, standing before cameras and microphones, one of the most powerful men in the technology industry pointed at something every family in the developed world has tucked behind a door somewhere, and said: this is how the next fifty years will work.
The argument, stripped to its bones, is this. The washing machine killed the laundromat as a profession. It eliminated a category of work that had employed real people, absorbed real hours, and sustained real livelihoods. Nobody mourned it publicly, because the machine arrived quietly, and the jobs drained away slowly. But in place of that lost profession, something else emerged — something nobody in 1920 could have named. The consumer appliances industry. The industrial designer. The refrigerator repairman. The electrician who wired the new kitchen. The shop assistant who sold the new machine. The marketing copywriter who described its virtues. The journalist who reviewed it. Entire occupational categories that did not have names before the machine made them necessary.
Huang's point is not a prediction. It is a pattern — one that, he argues, runs 150 years deep, from the earliest stirrings of industrial mechanisation to the present moment. "Just go back and look at history," he says. And so we shall.
But we shall also look at what history sometimes fails to tell us. Because the washing machine that frees you from laundry is not quite the same as the machine that begins to think.
II. The Pattern: One Hundred and Fifty Years of Consistent Evidence
The fear that machines will take all the jobs is not new. It is, in fact, one of the oldest recurring anxieties in industrial societies. The Luddites were not simply vandals; they were skilled textile workers who understood, with remarkable precision, that the machines being installed in the mills of early nineteenth-century England were designed to replace exactly what they did. They were right about the displacement. They were wrong about the destination.
Economists have spent considerable energy formalising why they were wrong. The most elegant formulation is also the most counterintuitive: the assumption that there is a fixed amount of work to be done — that when a machine takes a job, a human goes without — is simply false. It is known as the lump-of-labour fallacy, and it rests on a zero-sum view of the economy that economic history has repeatedly refuted. The demand for labour is not fixed. As one industry contracts, another expands. As one type of task disappears, new categories of human need emerge to replace it.
The 20th century is the strongest available evidence for this view. It was the most technologically disruptive century in recorded history — the century of electrification, the internal combustion engine, the telephone, the aeroplane, the computer, and the internet. It was also, across its entire arc, a century of expanding employment, rising wages, and increasing standards of living in the industrialised world. Unemployment did not structurally increase. The employment-to-population ratio did not permanently fall. The labour markets of developed economies absorbed wave after wave of technological disruption and still produced net job creation.
Economists Daron Acemoglu and Pascual Restrepo have given this dynamic its most precise modern description: the displacement effect and the reinstatement effect. New technologies displace workers from existing tasks. But they also reinstate demand for human labour in new tasks — often tasks that could not have been imagined before the technology arrived. Their research suggests that during the early phase of the Industrial Revolution, the displacement effect dominated, causing genuine suffering; but across the 20th century, reinstatement became the stronger force, driving up wages and living standards with it.
Robert Solow, the MIT economist who won the Nobel Prize in 1987 for defining the role of innovation in economic growth, once quipped that "you can see the computer age everywhere but in the productivity statistics." His paradox — that the most dramatic technological investment of the late 20th century seemed, initially, to produce very little measurable productivity gain — turned out to be an argument not against technology, but for patience. The gains came. They always come. But they come slowly, after the economy has had time to reorganise around the new tools, to train its workers, to restructure its institutions, and to discover the new forms of demand that the technology makes possible.
Jensen Huang knows this history. It is the foundation of his optimism.
III. The Internet's Unfinished Lesson
Huang's favoured modern example is the internet. It was supposed to kill retail. Instead, he argues, it created a five-trillion-dollar ecosystem that employs more people than bricks-and-mortar commerce ever did.
This is true at the aggregate level. It is also, at the human level, significantly incomplete.
The internet did kill large swaths of retail employment — and particularly in Europe, the numbers are not easily dismissed. According to the Euler Hermes analysis of European retail markets, for every job created in e-commerce, approximately four and a half jobs were lost by traditional retailers. European retail has been in what analysts have described as a permacrisis for more than a decade: rapid debt-fuelled expansion in the 2000s, the financial collapse of 2008, and then the steady erosion of footfall by e-commerce. In the United Kingdom alone, some 70,000 retail jobs were shed in the final months of 2018. Across Europe, front-end retail employment — the customer-facing roles, the shop assistants, the cashiers, the sales staff — declined measurably in every EU country between 2015 and 2022, with drops exceeding ten percentage points in Estonia, France, and Luxembourg.
These are not statistics in the abstract. They are the working lives of predominantly female employees in provincial towns, in city high streets, in the regional retail hubs that structured local economies for generations. The e-commerce jobs that replaced them — the warehouse operative, the logistics driver, the fulfilment centre worker — came with different geographies, different hours, and different bodies. They often appeared in places that had never previously had much employment at all, while vanishing from places that had.
The pattern held, in aggregate. History's reassurance was technically accurate. But it was accurate the way a doctor is accurate when she tells a patient with a broken leg that the bone will eventually heal: true, necessary, and wholly insufficient as a response to the present pain.
This is the first crack in the optimist's certainty. The historical pattern does not protect specific communities, specific skills, or specific timescales. It does not guarantee that the new jobs will appear in the same towns, require the same people, or arrive within a politically survivable window. The transition period — the gap between the displacement and the reinstatement — is not a rounding error. It is, for the people who live through it, the whole story.
IV. The Qualitative Break: When the Machine Learns to Think
Every previous technology that displaced human labour shared a crucial characteristic: it replaced a specific, bounded, physically or arithmetically defined capability. The steam engine replaced muscle. The mechanical loom replaced weaving fingers. The electronic calculator replaced the human calculator. The automated switchboard replaced the telephone operator. In each case, the human worker was displaced from a particular task — but retained everything else. The general cognitive apparatus, the relational intelligence, the capacity for judgement, for creativity, for learning new tasks: all of this remained human property, untouched by the machine.
This is why the historical pattern of reinstatement held. Workers could move from the displaced task to an adjacent one, or a new one, carrying their general human capabilities with them. The calculator couldn't follow them. The loom couldn't chase them into the new occupations they created. The machines stayed in their lane.
Artificial intelligence does not stay in its lane.
This is not a matter of sentiment or fear. It is a structural observation about what AI is designed to do. Unlike every previous automation technology, AI is not being built to perform a specific task more efficiently than humans. It is being built to replicate the general cognitive capacity of humans — to understand language, to reason, to adapt to novel circumstances, to produce outputs across an open-ended range of domains. Research into the labour market impacts of AI has been explicit on this point: AI differs qualitatively from other automation technologies because it is not task-specific. It possesses general cognitive skills, and an AI capable of automating a wide range of current tasks without overfitting may also be capable of automating new tasks as they emerge.
This breaks the historical model in a specific and important way. When a previous technology displaced workers from Task A, those workers could retrain for Task B. The technology could not follow them into Task B, because it was designed for Task A. But if AI has general cognitive capability, it can potentially follow workers from Task A to Task B, and from Task B to Task C, because what it possesses is not a solution to a specific problem but a generalised problem-solving capacity.
There is a further dimension that Huang's washing machine analogy does not capture: the network embeddedness of AI. Previous machines required humans to remain present — to operate them, to supervise them, to make decisions at the boundary between the machine's capability and the world's complexity. The washing machine required someone to load and unload it. The assembly line required human hands at the joints, human judgement at the quality checkpoints, human intelligence at the design stage. Humans remained structurally necessary at each node of the production process.
AI is beginning to operate differently. It is being embedded into operational network systems — logistics platforms, financial systems, customer service infrastructures, medical diagnostic pipelines, legal research tools — in ways that progressively reduce the number of decision nodes at which a human presence is required. The task does not merely proceed faster. It proceeds, increasingly, without the human needing to be present at all — not physically, and in many cases not even mentally. The AI interacts with the system, consults the data, produces the output, and delivers the result. The human, at the far end, reviews a summary.
This is not acceleration. This is substitution of a different order.
V. The Cognitive Atrophy Argument
There is a dimension to AI's integration into work that the economic models do not yet fully capture, and it concerns not employment but cognition itself.
When a previous generation of technology accelerated human productive capacity, it did so without altering the human cognitive apparatus. The person who used an electric drill was still obliged to understand woodwork. The person who used a spreadsheet was still obliged to understand the numbers. The tool extended the reach of competence; it did not replace the need for competence.
AI begins to change this relationship. Research on AI-assisted workers has found that while AI assistance produces significantly greater task-level efficiency — higher accuracy, faster completion — it does not produce differential gains in underlying cognitive ability. Workers assisted by AI perform the task better. They do not become better thinkers. And the mechanism that explains this finding is well established in cognitive science: learning and adaptive expertise depend on sustained cognitive effort and the active deployment of internal resources. When the effort is outsourced to the machine, the skill does not deepen. It may, over time, atrophy.
The washing machine freed people from believing that laundry was all they were built for. But it did not reduce their capacity to do other things. It freed up time and energy that could be redirected into new human activities. AI risks doing something subtly different: freeing people from the cognitive effort of certain tasks, while also, over time, reducing their capacity to perform those tasks independently. The liberation and the diminishment arrive together.
This matters for the reinstatement argument because the historical mechanism of reinstatement depended on human cognitive plasticity: the ability to learn new skills, to adapt to new occupational demands, to invent new categories of work. If AI's pervasive integration into daily cognitive life gradually reduces that plasticity — if the cognitive muscles required for adaptation are the very ones that AI use progressively underemploys — then the reinstatement mechanism is impaired at its root.
VI. Where Huang Is Right — And the Limits of the Objection
It would be intellectually dishonest to marshal the counterarguments without acknowledging their limits, and without giving Huang's optimism its proper weight.
The radiologist case is among the most cited and most instructive. AI has become a powerful tool in radiological image analysis, capable of identifying anomalies with speed and consistency that no human practitioner can match. By the logic of displacement, radiology should be contracting. Instead, there are more radiologists now than before AI arrived. The explanation is consistent with the historical pattern: AI automation of certain tasks within a profession does not eliminate the profession if it simultaneously expands the scope of what the profession can accomplish. AI does not replace the radiologist; it makes the radiologist more productive, allowing more scans to be analysed, more conditions to be monitored, more patients to be served.
The same dynamic has been observed in hospitals more broadly. As AI handles administrative burden and diagnostic assistance, hospitals have found themselves able to treat more patients — and to hire more nurses. The AI-enhanced hospital is a bigger hospital, not a smaller one.
NVIDIA's own investment data supports a broader version of this argument. The year 2025 was the largest year for global venture capital investment on record, with more than a hundred billion dollars deployed worldwide, the majority of it into AI-native companies spanning healthcare, robotics, manufacturing, and financial services. That investment is already translating into concrete job creation: demand for data centre construction workers, network technicians, electricians, and steelworkers — the unglamorous physical infrastructure of the digital economy. The PwC 2025 Global AI Employment Barometer found that workers with AI skills commanded an average wage premium of 56 per cent — twice the figure of a year earlier. Even in roles most directly affected by AI, employment rose 38 per cent.
These are not trivial numbers. They suggest that the historical pattern of reinstatement is at work, visibly and measurably, in the early stages of the AI transition.
And yet. The optimist's one honest caveat — Huang's own — is also the hinge on which everything turns. "If the world runs out of ideas," he told CNN's Fareed Zakaria, "then productivity gains translate to job loss." This is not a minor qualification. It is the entire argument. Every historical pattern of job reinstatement rested on the human capacity to generate enough new wants, new industries, and new imaginative demands to absorb the displaced. The unprecedented question that AI poses is whether the generator of new ideas can itself be partially automated — and whether that automation arrives before the new ideas do.
VII. The Threshold Problem: When Displacement Outruns Reinstatement
The most sobering academic treatment of this question does not argue that AI will inevitably destroy more jobs than it creates. It argues something more precise and more troubling: that there may exist a threshold — an AI-capital-to-labour ratio — beyond which the reinstatement mechanism breaks down. Below the threshold, the historical pattern holds: displacement creates disruption, reinstatement follows, the economy adapts. Above the threshold, a self-reinforcing cycle of recessionary pressures could be triggered, exacerbating social disparities, reducing social cohesion, heightening political tensions, and requiring sustained government intervention to maintain stability.
The question is not whether we will reach that threshold. The question is whether we will recognise it before or after we have crossed it.
Previous technological transitions moved at human institutional speed. The Industrial Revolution unfolded across decades, giving parliaments, trade unions, and social reformers time — painful, insufficient, often agonisingly slow time — to respond. The internet's retail disruption unfolded across fifteen years. AI's penetration of cognitive labour markets may unfold across five years, or three, or less. The speed of transformation is itself a variable that historical analogies do not adequately address.
There is also the combinatorial problem. Previous technological transitions targeted either physical labour or cognitive labour, rarely both simultaneously. The assembly line threatened the factory worker. The spreadsheet threatened the accountant. AI, in conjunction with advancing robotics, threatens both at once — and it does so in an interlocking way, because the cognitive automation of planning and management reduces the oversight requirements for physical automation, which in turn reduces the demand for cognitive labour managing physical systems. When AI and robotics converge, the traditional retraining corridor — the human path from manual work to cognitive work that carried generations through previous transitions — begins to narrow.
VIII. The Civilisational Question
The deepest dimension of this transition is not economic. It is anthropological.
Every previous technology that displaced human labour did so at the periphery of human identity. The person who could no longer earn a living by weaving could, in principle, redirect their human capacities — their relational intelligence, their creativity, their embodied skill — into new forms of productive activity. Their humanness remained intact as an economic resource. The machine displaced the task. The human retained the territory.
AI's challenge is more intimate. By targeting not just physical tasks but cognitive ones — by demonstrating competence in reasoning, language, analysis, and increasingly in creative synthesis — AI does not merely displace tasks at the periphery of human identity. It moves toward the centre. Not, perhaps, all the way to the centre. There remain dimensions of human presence, judgement, and relationality that resist automation — the Moravec gap, as researchers have called it, between the tasks that AI finds easy (mathematical reasoning, language processing, pattern recognition) and those it finds hard (dexterous physical manipulation, real-time sensorimotor adaptation, genuine empathic presence).
But the boundary is moving. And the civilisational question that no economic model can answer is this: if the things that made human beings economically necessary are progressively automated, where do human beings locate their sense of purpose, their sense of worth, and their claim to participation in the collective project?
The washing machine didn't answer this question. It dissolved it, because the time it freed was filled immediately by new economic necessity. The historical pattern worked because new needs always appeared fast enough to absorb the liberated energy. AI may work the same way. The new needs may appear, the new industries may emerge, and the transition — painful, uneven, geographically unjust — may ultimately resolve in humanity's favour, as it has always done before.
Or it may not. For the first time in the history of technology, we are in the presence of a machine that does not merely help us go faster. It begins, in some domains, to think on our behalf. And we have no established template for what follows.
IX. What Huang Got Right, What He Left Out, and What We Must Do With Both
Jensen Huang is not wrong. He is incomplete.
The 150-year pattern he invokes is real. The lump-of-labour fallacy is a genuine fallacy. The washing machine really did generate more economic activity than it destroyed. The internet really did create more jobs than it killed — in aggregate, over time, across the whole system. The radiologist really is more employed than before AI arrived.
But the internet also restructured European retail in ways that caused genuine, lasting harm to specific communities — and the compensation arrived late, unevenly, and in different places from the damage. AI, operating simultaneously on cognitive and physical labour, at digital speed, with the capacity to embed itself into network systems and reduce the requirement for human presence at cognitive nodes, is a different order of challenge from anything that came before it.
The honest position is not optimism or pessimism. It is precision. The historical pattern probably holds — but at the aggregate level, over time, with enormous human cost in the transition. The reinstatement mechanism probably operates — but it depends on human cognitive plasticity, on institutional speed, on political will, and on the continued human capacity to generate new imaginative demands faster than AI can satisfy the old ones.
These are not certainties. They are conditions. And whether those conditions are met is not a technological question. It is a political, educational, institutional, and above all a civilisational one.
The machine washes the clothes. What we do with the free time — whether we build new lives or simply watch the machine run — has never been determined by the machine itself.
Ten Questions Worth Carrying From This Article
1. Is it true that technology has always created more jobs than it destroyed?At the aggregate level, over sufficient time: yes. The 20th century — the most disruptive technological century in history — produced net job creation and rising wages across the industrialised world. The pattern is real. The suffering in the transition was also real, and the pattern provides no comfort to those who live through the gap.
2. Did the internet actually create more jobs than it killed in retail?In aggregate, yes — but the distribution was deeply uneven. For every e-commerce job created in Europe, approximately four and a half traditional retail jobs were lost. The new jobs appeared in logistics hubs and digital platforms; the lost jobs disappeared from high streets and provincial town centres. The net is positive; the geography is unjust.
3. What makes AI genuinely different from previous automation technologies?Every previous machine replaced a specific, bounded capability. AI is designed to replicate the general cognitive capacity of humans — the capacity to reason, to adapt, and to learn new tasks. This means that the historical escape route — workers moving from automated tasks to new ones — is not guaranteed, because AI may be capable of following them.
4. What is Jensen Huang's own honest caveat about the optimist's case?He stated it plainly: "If the world runs out of ideas, then productivity gains translate to job loss." The entire historical pattern rests on humanity's capacity to generate new wants and new industries faster than technology eliminates the old ones. AI's unique challenge is whether it can automate the generation of ideas before the new ideas arrive.
5. What is the "threshold argument" in AI employment research?Some economists argue that there may exist a ratio of AI capital to human labour beyond which the reinstatement mechanism breaks down — triggering a self-reinforcing cycle of displacement, social fragmentation, and political instability that requires sustained government intervention to stabilise. We do not know where that threshold is.
6. How does AI risk changing not just what we do but how we think?Research has found that AI assistance improves task-level performance without improving underlying cognitive ability. Workers who use AI become better at the task; they do not become better thinkers. Over time, cognitive skills not exercised may atrophy — and it is precisely those skills of independent reasoning and adaptation that the reinstatement mechanism historically depended upon.
7. Is AI + robotics converging in a way that is qualitatively new?Yes. Previous transitions targeted either physical or cognitive labour. AI and advanced robotics are converging simultaneously, closing the traditional retraining corridor — the human path from manual to cognitive work — that carried earlier generations through displacement.
8. What does "embedding AI into network systems" actually mean for workers?It means that AI is being integrated into operational processes — logistics, finance, customer service, diagnostics — in ways that progressively reduce the number of decision points at which human presence is required. The task proceeds without the human needing to be present, mentally or physically, at intermediate stages.
9. What can we learn from the radiologist case?That AI automation of specific tasks within a profession does not automatically eliminate the profession if it simultaneously expands the scope of what that profession can accomplish. AI-enhanced radiologists see more patients, not fewer. The lesson: what matters is whether AI amplifies the professional's purpose or replaces it.
10. If the pattern probably holds, why should we be concerned at all?Because "probably holds in aggregate over time" is not the same as "will hold humanely, in the right places, within a politically stable timeframe." The transition period — the gap between displacement and reinstatement — can be long enough to cause democratic erosion, community collapse, and generational immiseration, even if the eventual destination is positive.
Author's Note: Limits, Tensions, Lizards, and the Tools of Argument
On the limits of this article
This piece is built on the collision between a genuine pattern and a genuine exception. I have tried to honour both without resolving the tension prematurely, because I believe the tension is real and the resolution is not yet available to any honest observer. There are things I do not know, and two of them are large: whether the AI transition will ultimately follow the historical pattern of reinstatement, and how long the transition period will be. Anyone who claims certainty on either point is selling something.
The data on European retail job losses is real but imprecise — the "three million" figure cited in the framing of this article represents a broad regional estimate across multiple sectors and timeframes, not a single peer-reviewed count. I have used corroborating data from OECD, Euler Hermes, and Centre for Retail Research to triangulate the direction and scale of the phenomenon without overstating a specific number. Readers who want precision should consult those primary sources.
On subjective biases
I am a person who earns his living in the digital economy, who has spent years helping Romanian public institutions build their online presence, and who has watched the disruption of traditional service markets from the inside. I am, therefore, neither a detached academic nor a disinterested observer. I am inclined toward the view that technology requires institutional response — that the market alone will not distribute the gains of AI equitably, and that the communities most at risk are those with the least political voice. This inclination probably colours the emphasis of this piece, and readers should weigh it accordingly.
On the lizards: Aesopic language decoded
The washing machine in this article is a lizard. Not literally — but rhetorically. Huang chose it precisely because it is harmless, familiar, and domestic. A washing machine cannot possibly threaten anyone. By anchoring his argument about one of the most consequential technological transitions in human history to an appliance found in every home, he disarms the anxiety before the argument begins. It is a rhetorical act of normalisation — and it is entirely deliberate.
The lizard I have introduced in response is the "cognitive corridor" — the idea that humans have always been able to move from automated tasks to new ones, carrying their general intelligence with them. This corridor is the hidden structural assumption in every optimist's reassurance, and I have tried to make it visible, because the question of whether AI narrows or closes that corridor is the most important unanswered question in this entire debate.
Watch for lizards in arguments about AI. They usually appear as reassurances that are technically true but structurally incomplete: "AI will create more jobs" (true, probably — but where, when, and for whom?); "there will always be things humans do better" (true, currently — but the boundary is moving); "history always proved the pessimists wrong" (true — but history also included enormous human suffering in the transitions that eventually resolved well).
On the rhetorical tools used in this article
The article uses the following devices consciously:
Anaphora — the repeated return to the washing machine as anchor, each time carrying slightly more weight than the last, so that by the final paragraph it has accumulated the full argument within a single domestic image.
Antithesis — the structure throughout is built on paired oppositions: pattern and exception, displacement and reinstatement, aggregate and local, speed and time. The truth lives in the tension, not in either pole.
Concession before extension — I have tried consistently to give Huang's argument its strongest form before extending or qualifying it. This is not rhetorical generosity; it is epistemological hygiene. An argument that only engages with weak versions of its opposition is not an argument about the world; it is a performance of certainty.
The delayed thesis — the article's actual argument is not stated until Section IX, because the preceding sections are necessary to make the argument honest. Readers who skip to the conclusion will find a position; readers who follow the full arc will understand why that position is held with the specific qualifications it carries.
Call to Action: What Can You Do Tomorrow?
The analysis above is not a counsel of despair. It is a counsel of precision. Here are five concrete actions worth taking in the days immediately ahead.
1. Map your task profile, not your job title. Your job title is a social category. Your task profile is what you actually do each day. Write down the ten to fifteen recurring tasks that constitute your working week. Then ask, honestly, which of those tasks are fundamentally about cognitive output that could be replicated — and which are about human presence, relational judgement, or contextual wisdom that cannot. The gap between those two lists is where your irreplaceable value currently lives. Protect it. Develop it. Do not let it atrophy through over-reliance on AI assistance.
2. Invest in the tasks AI finds hardest. Research into the computational cost of different forms of automation (the Moravec gap) suggests that dexterous physical manipulation, real-time relational judgement, genuine empathic presence, and contextually embedded decision-making remain extraordinarily difficult for AI. These are not soft skills. They are hard skills of a different kind. Invest in them as deliberately as you would invest in technical proficiency.
3. Follow the geography of the new investment, not just the headline. AI is generating enormous amounts of new economic activity. The question is where. Track where the new AI-native companies are being built, where the data centres are being constructed, where the new occupations are emerging. If those geographies do not include your community, that is actionable information — either for relocation, for political advocacy, or for community-level economic planning.
4. Maintain your independent cognitive practice. Use AI as a tool, not as a replacement for thinking. Preserve the habits of sustained reading, original analysis, and unassisted problem-solving that constitute cognitive resilience. The research is clear: AI assistance improves performance; it does not improve the performer. You need the performer to remain capable, because the machine can malfunction, the subscription can lapse, and the task may one day require you to think without assistance.
5. Engage with the institutional question, not just the personal one. The most important responses to AI displacement are not individual. They are collective: education systems that prioritise adaptability over specific technical skills; social safety nets that are designed for people in transition rather than people in stable employment; democratic institutions that can respond to technological disruption at the speed at which it occurs. These require political engagement, not just professional adaptation. Know which politicians in your constituency are thinking seriously about these questions, and hold them to account.
What to avoid:
Do not panic-retrain into a specific technical skill assuming it will be AI-proof. The half-life of technical AI-adjacent skills is shortening rapidly. Flexibility of mind is more durable than any specific certification.
Do not dismiss the concern because the historical pattern is reassuring. The historical pattern is real; it is also an average, and averages conceal enormous human variation. Your community, your sector, and your timescale may be in the painful half of the distribution.
Do not outsource your entire professional identity to AI tools before you understand what you are outsourcing. The thing you hand to the machine is not just a task. It may be the cognitive practice that made you capable of doing the task in the first place.
Do not wait for certainty before acting. The transition is already underway. The radiologist who waits to understand AI before engaging with it will be served less well than the radiologist who engages with it before fully understanding it.
Do not mistake the burial of a task for the burial of a person. This is Huang's most important insight, and it is correct. The work you did yesterday is not the same as who you are. The machine that makes yesterday's work obsolete does not make you obsolete — unless you have spent so long identifying with the task that you have forgotten you are more than it.
Petru Cojocaru is a writer and digital strategist based in Botoșani, Romania. He writes on technology, civilisation, and the slow edge between them at petrucojocaru.ro
Bibliography and Sources
- Jensen Huang, Fox News interview, June 2026
- Jensen Huang, World Economic Forum, Davos, January 2026
- Jensen Huang, Carnegie Mellon University Commencement Address, May 2026
- Jensen Huang, Fortune / IEEE interview, April 2026
- Jensen Huang, CNN / Fareed Zakaria interview, 2025
- Daron Acemoglu & Pascual Restrepo, "Automation and New Tasks: How Technology Displaces and Reinstates Labour", Journal of Economic Perspectives, 2019
- David Autor, "Why Are There Still So Many Jobs?", Journal of Economic Perspectives, 2015
- David Autor et al., "New Frontiers: The Origins and Content of New Work, 1940–2018", Quarterly Journal of Economics, 2024
- Robert Gordon, The Rise and Fall of American Growth, Princeton University Press, 2016
- Niall Kishtainy, "A New Industrial Revolution?", IMF Finance & Development, December 2025
- OECD, Local Retail, Global Trends: Addressing Labour and Skills Shortages in Europe's Retail Sector, January 2026
- Euler Hermes, European retail displacement study (referenced via Retail Dive, January 2020)
- Centre for Retail Research, The Crisis in Retailing, ongoing
- British Retail Consortium, retail employment surveys, 2018–2019
- PwC, 2025 Global AI Employment Barometer
- Erik Brynjolfsson, "The Productivity Paradox", MIT Technology Review
- arXiv 2510.26787, Remote Labor Index: Measuring AI Automation of Remote Work
- arXiv 2506.10281, Closer to Language than Steam: AI as the Cognitive Engine of a New Productivity Revolution
- arXiv 2403.17405, The Recessionary Pressures of Generative AI: A Threat to Wellbeing
- arXiv 2509.24466, Moravec's Paradox and Restrepo's Model: Limits of AGI Automation in Growth
- arXiv 2510.24893, Efficiency Without Cognitive Change: Evidence from Human Interaction with Narrow AI Systems
- St. Louis Federal Reserve, "Examining the Lump of Labour Fallacy", 2020
- NVIDIA Blog, Davos WEF coverage, January 2026