Adam Smith’s pin factory is perhaps the most famous thought experiment in economics. By dividing labour into eighteen distinct operations, Smith observed, ten workers could produce 48,000 pins per day rather than the handful each might manage working alone. This observation became the foundation of modern economic thinking about productivity: specialisation is how we get richer.
The story is so deeply embedded in how we think about economic progress that it feels almost definitional. Developed economies are specialised economies. Rich countries have workers with narrow, deep expertise. Poor countries have generalists scratching out a subsistence living. The path from the latter to the former runs through ever-finer divisions of labour.
This story is incomplete. Technology sometimes works in exactly the opposite direction, and we’re about to see one of the largest despecialisation events in economic history play out in the software industry.
Consider the sewing machine. Before its widespread adoption in the mid-19th century, clothing production was a specialised trade involving cutters, tailors, seamstresses, and finishers. The home sewing machine allowed a single person to produce entire garments, despecialising clothing production for household and small-scale use even as industrial garment factories moved in the opposite direction.
The tractor tells a similar story. Pre-mechanised farming often involved specialists in ploughing, harvesting, threshing, and animal husbandry. A farmer with a tractor could perform the work of several specialised roles, contributing to massive rural depopulation as generalist owner-operators replaced larger teams. American agricultural employment fell from roughly 40% of the workforce in 1900 to under 2% today, while output increased dramatically.
The automobile eliminated not just horses but an entire ecosystem of specialists: farriers, stable hands, harness makers, carriage builders, and wheelwrights. The telephone collapsed the roles of messengers, telegraph operators, and certain clerical positions into direct communication. Desktop publishing eliminated typesetters, layout artists, and paste-up technicians in favour of one person with PageMaker.
These aren’t obscure edge cases. They’re some of the most significant technological transitions of the past two centuries, and they all involved substantial despecialisation of the workforce.
There’s a common thread running through these examples: despecialisation happens when a technology encapsulates complexity, hiding specialist knowledge inside a tool so that the user doesn’t need to possess it themselves.
The calculator encapsulates the clerk’s arithmetic skill and knowledge of computational shortcuts. The word processor with spell check encapsulates the proofreader’s knowledge of spelling and grammar conventions. GPS navigation encapsulates the taxi driver’s hard-won knowledge of city streets. In each case, expertise that took years to acquire becomes accessible to anyone who can operate the tool.
Sufficiently powerful tools don’t just augment specialists; they can replace the need for specialisation entirely. The knowledge doesn’t disappear. It migrates into the tool, and whoever can operate the tool gains access to capabilities that previously required dedicated training. This is the mechanism by which technology produces generalists rather than specialists, and it’s been happening for centuries alongside the more celebrated story of increasing division of labour.
Of course, specialisation doesn’t disappear entirely. When a technology despecialises its users, it simultaneously creates new specialists upstream: the engineers, designers, and manufacturers who build the tool.
The ratios tell a different story, though. A few thousand people at Apple enable hundreds of millions of users to be their own photographer, navigator, secretary, and translator. The specialists building the technology are vastly outnumbered by the generalists using it. The lived experience of work for the overwhelming majority of people touched by these technologies is despecialisation, even if somewhere a small team became hyper-specialised to make it possible.
Economies can become more specialised in producing technologies while simultaneously becoming less specialised in using them. The specialisation is real, but it concentrates in an ever-smaller fraction of the workforce while the majority becomes more generalist.
The current structure of software teams is highly specialised. A typical product team at a well-funded startup might include frontend engineers, backend engineers, infrastructure specialists, a product manager, a designer, a QA engineer, and a data analyst. Much of the coordination overhead in software development exists purely to manage handoffs between these roles: meetings, tickets, specifications, design reviews, code reviews, deployment approvals.
If AI tools become sufficiently capable at each of these individual tasks, a single person with strong product judgment could sketch a UI and have it translated to working code, write backend logic with AI assistance across unfamiliar frameworks, spin up infrastructure without deep cloud knowledge, query and analyse data without writing SQL from scratch, and generate and validate tests automatically.
The binding constraint shifts from “can you execute this specialised task” to “do you know what’s worth building and why.” This is a product sense and judgment role more than a technical specialisation. Call it a “Product Engineer” rather than a software engineer.
The skill profile looks quite different from what current software engineer hiring selects for. It emphasises comfort with ambiguity, the ability to understand technical complexity without necessarily implementing it yourself, strong intuition about what people need, and good taste in what constitutes an elegant solution. This is almost orthogonal to what gets rewarded in technical interviews, which heavily emphasise algorithmic problem-solving, language-specific knowledge, and system design within established patterns. The Product Engineer sounds more like a startup founder than a typical senior individual contributor.
The transition may be brutal for some current specialists. Designers face a particularly difficult outlook: if AI can generate UI variations quickly and most software just needs clear, functional, reasonably attractive interfaces rather than exceptional design, the taste component gets absorbed into the Product Engineer role. What remains for dedicated designers is the very high end, and that’s a small fraction of current design employment.
Engineers who optimised for deep technical specialisation face a similar squeeze. The industry selected heavily for people who are comfortable with well-defined problems, who enjoy the craft of implementation, who might actively dislike ambiguity and stakeholder management. Those people optimised for a game whose rules are changing underneath them. Meanwhile, someone dismissed as “not technical enough” to be a real engineer but with strong product instincts might suddenly find themselves well-positioned.
What happens to compensation? Probably something like: fewer people, doing more, paid more per person, but with a lower total labour share of the value created (with the remainder going to capital). The industry shrinks in headcount while output increases. Individual compensation rises for those who successfully transition, but a substantial portion of current software workers might find themselves displaced rather than elevated.
The despecialisation thesis doesn’t overturn Smith entirely. It suggests that specialisation and despecialisation operate in cycles: each wave of specialisation creates complexity and coordination costs, generating demand for integrating technologies that enable a new generation of generalists, who then discover new opportunities for specialisation at a higher level of abstraction. The direction of travel isn’t a linear march toward ever-finer divisions of labour, but something more like a spiral.
The software industry might be approaching the end of one such cycle, about to collapse decades of accumulated specialisation into a new generalist role that would have been impossible without the underlying AI technology. If this is right, the winners won’t be those who double down on their current specialisations, but those who position themselves for the coming integration: building the judgment, taste, and comfort with ambiguity that will matter when implementation stops being the bottleneck. That means fully embracing AI tools rather than resisting them, and expanding into adjacent skills rather than deepening existing ones.
The pin factory made specialisation famous, but the sewing machine should have made us pay more attention to the opposite force. We might be about to get a very expensive lesson in what happens when we forget half the story.