SWITCHED ON
The daily technology series nobody asked for but everyone needed
Your Job, The Algorithm, and What Comes Next
Every previous wave of automation created more jobs than it destroyed. The people who believe AI is categorically different this time are not alarmists. They have an argument worth taking seriously.
Previous waves of automation replaced muscle. The loom replaced the weaver's hands. The tractor replaced the farmhand's back. The assembly robot replaced the factory floor. AI is the first automation wave that credibly threatens to replace the activities that happen between the ears. That is not the same problem, and the historical reassurance that new jobs always emerge may not apply in the same way.
Yesterday we went deep into semiconductor geopolitics — TSMC's extraordinary concentration in Taiwan, ASML's monopoly on the machines that make leading-edge chips possible, the US Chips Act and its slower-than-advertised reshoring, China's export-control-constrained domestic ambitions, and NVIDIA's position as the most strategically important company in the global AI economy. Today we are zooming from the hardware layer to the human layer, and asking the question that sits underneath every episode of this series since Episode One: what does all of this do to work? Not in the abstract future-of-work-thought-leadership sense that fills LinkedIn with content that sounds profound and means nothing. In the specific, documented, already-happening sense. What jobs are being automated now. What jobs are next. Why this technological wave might be different from the ones that preceded it. And what, if anything, the people responsible for economic policy are actually doing about it.
01 — The Historical Reassurance and Why It May Not Hold
The standard response to automation anxiety is historical. Every previous wave of labour-displacing technology — the spinning jenny, the steam engine, electrification, the computer — produced short-term displacement and long-term job creation. The jobs that were destroyed were replaced by new jobs that did not previously exist. The agricultural revolution did not produce mass permanent unemployment; it freed labour for the industrial economy. Industrial automation did not produce mass permanent unemployment; it freed labour for the service economy. The optimists argue that AI will follow the same pattern, creating new categories of work we cannot currently imagine while eliminating categories we can, and that the net effect over a generation will be positive.
This argument is not wrong about the history. Every previous displacement did eventually produce new employment. The question is whether the mechanism that drove that outcome — the creation of new tasks requiring human labour faster than old tasks were automated — will function the same way when the technology being deployed is capable of performing cognitive tasks. Previous automation was tool-like: it amplified human capability in specific physical domains while remaining entirely dependent on human cognition for direction, planning, problem-solving, and judgment. AI is the first technology that automates cognition itself. The new tasks that previous automation created were cognitive ones. If the technology that automates those cognitive tasks can also automate the new cognitive tasks it creates, the self-correcting mechanism of the historical record faces a challenge it has not previously encountered.
The weavers displaced by the loom became factory supervisors. The factory workers displaced by robots became software engineers. The software engineers displaced by AI become — what, exactly? The historical answer of "something we can't imagine yet" is less reassuring when the technology doing the displacing can increasingly do the imagining too.
02 — What Is Actually Happening Right Now
The displacement is not coming. It is already underway, unevenly distributed, and considerably less visible than the announcements about AI capabilities would suggest — partly because economic change happens slowly even when technology changes fast, and partly because much of the current impact is in reduced hiring rather than direct layoffs, which shows up in employment statistics with a lag.
The clearest documented impacts are in specific cognitive task categories. Legal research, once performed by junior associates billing at high hourly rates, is now substantially automated at large firms using AI document review and legal research tools. The demand for entry-level legal research work has fallen significantly. Customer service operations have automated first-contact handling at scale — not always to customer satisfaction, but effectively enough to reduce headcount. Radiologists reading routine scans, junior software developers writing boilerplate code, copywriters producing templated content, financial analysts running standard reports — in each of these categories, the volume of human work required per unit of output has fallen measurably in organisations that have adopted AI tools.
Crucially, these are not the jobs that previous automation waves targeted. Previous automation primarily displaced manual labour in manufacturing, agriculture, and logistics. The jobs being disrupted now are predominantly white-collar, require significant education and training, and sit in the middle of the income distribution. The doctor, lawyer, accountant, and engineer — the professions that represented the secure, educated, middle-class pathway that post-industrial economies promised — are the professions facing the most immediate AI disruption, not because AI is superhuman in these domains but because it is good enough at the routine portions of these jobs to change the economics of performing them with human labour.
03 — The Geography of Disruption
The impact of AI-driven labour displacement is not distributed evenly across geography, income level, or education. This matters enormously for the political and policy response, because disruption that concentrates in specific communities produces political consequences very different from disruption that distributes broadly across the population.
Emerging economies that were pursuing a development pathway through exported services — call centres, software development, data annotation, back-office processing — face a particular challenge. India's IT services sector, which employs millions of workers and has been a significant driver of middle-class expansion, is heavily exposed to AI automation of the tasks it has historically performed. The development pathway that allowed countries like India, the Philippines, and several others to lift workers into stable employment by serving the global demand for cognitive outsourcing is being compressed by the same AI capabilities being developed primarily in the countries that previously outsourced the work.
Within wealthy economies, the disruption splits along lines that defy the usual political narratives about automation. The workers most exposed are not the lowest-income — physical labour in care, construction, and personal services is proving more durable than routine cognitive work, because AI is currently better at pattern recognition in well-defined domains than at the embodied judgment, dexterity, and interpersonal sensitivity that physical service work requires. The workers most exposed are in the middle: educated, salaried, performing the kinds of structured cognitive tasks — research, drafting, analysis, data processing — that AI performs most readily. This is politically volatile territory, because these are the workers who least expected to be disrupted and have the most resources to make their displeasure felt.
04 — The Policy Responses, Graded Honestly
The policy conversation about AI and work ranges from serious to performative, and distinguishing between them requires some patience with the gap between what politicians say and what the evidence supports.
Reskilling and retraining programs are the most common political response, and the evidence for their effectiveness at the scale required is mixed at best. Most reskilling programs work for some workers in some circumstances and fail to reach the workers who most need them — older workers with established careers and significant switching costs, workers in communities where the new jobs being created are geographically distant from the jobs being lost, and workers for whom the cognitive demands of reskilling for AI-resistant roles are genuinely difficult. Announcing a reskilling program is considerably easier than designing one that works and funding it at adequate scale. Most governments have done the former without adequately doing the latter.
Universal Basic Income — providing every citizen a regular unconditional cash payment regardless of employment status — has moved from fringe proposal to mainstream policy debate with notable speed. Pilots have been conducted in Finland, Kenya, Canada, and several US cities. The consistent finding is that recipients do not stop working, mental health improves, and entrepreneurial activity increases. The consistent objection is cost — a meaningful UBI for an entire adult population of a developed economy is extraordinarily expensive, and the funding mechanisms proposed are variously inadequate, politically toxic, or both. The debate has not resolved. It has intensified.
Shorter working weeks — the four-day week movement — have been piloted at company and national level, with Iceland's large-scale trial producing broadly positive productivity results that have been replicated in smaller studies elsewhere. The logic is that if AI increases productivity per worker-hour, the gains should be distributed partially as time rather than entirely as profit. This is a reasonable distributional argument that requires significant cultural and structural change to implement and has not yet found legislative traction at national scale in any major economy.
Every serious policy response to AI-driven labour displacement requires one thing none of the current proposals adequately addresses: the political will to tax the productivity gains from automation and redistribute them to the workers displaced by it. Everything else is rearranging the furniture.
05 — What Work Actually Becomes
The most intellectually honest framing of the future of work is not that jobs disappear but that the composition of what constitutes valuable human labour shifts, in ways that are already partially visible and partially opaque. Physical care work — nursing, childcare, elder care, personal support — is both AI-resistant and chronically undervalued by markets relative to its social importance. The question of whether societies will revalue this work when more easily automatable cognitive work is no longer the scarce resource it once was is genuinely open.
Human judgment in high-stakes, genuinely novel situations — the surgeon in an unexpected complication, the negotiator managing a relationship breakdown, the leader making a decision under genuine uncertainty with incomplete information — retains value that routine cognitive work does not, because it requires integrating embodied experience, interpersonal sensitivity, and contextual wisdom in ways current AI cannot replicate. The labour market implication is a polarisation between work that is genuinely irreducible to algorithm and work that is not, with the middle category — structured cognitive tasks performed by educated workers — squeezed from above by AI and below by cost pressure.
The optimistic version of this future is a world in which AI handles the drudgery and humans do the things that are most distinctively human — creative, relational, embodied, judicious. The pessimistic version is a world in which the economic value of distinctively human activities is insufficient to support the living standards that educated, middle-income populations in wealthy countries currently expect, and the political consequences of that gap are severe. Both versions are plausible. Which one materialises depends less on the technology than on the distributional choices made about who benefits from the productivity gains AI generates. Those choices are political, not technical, and they have barely begun to be made.
Tomorrow we are moving from the economy of the future to the politics of the present — specifically the intersection of technology and democracy. Disinformation at scale, deepfakes in electoral contexts, social media's documented effects on political polarisation, and the question of whether the information environment that democracies depend on can survive what AI is doing to it. See you then.
Switched On is a daily technology series covering the ideas, systems, and arguments shaping the digital world. Opinionated. Witty. Occasionally wrong. Always worth the argument.
