Why do we keep assuming the economy is a clean machine, when it behaves more like an old forest? New shoots appear, yes, but most of the trunk remains where it was. That is the uncomfortable lesson in a recent study of Swedish jobs: the future is not always built by invention alone. Sometimes it is built by division, by splitting tasks thinner and thinner until work that once sat inside one craft becomes many smaller crafts. In other words, the labor market may be less Schumpeter’s bonfire than Adam Smith’s sawmill.
The study, by William Skoglund, Jakob Molinder and Kerstin Enflo, uses full-count Swedish census and administrative data from 1880 to 2019 harmonized to five-digit HISCO. Its core finding is awkward for anyone who likes simple narratives about technological progress. More than 70% of employment today sits in occupations whose core economic functions already existed in the late nineteenth century. That does not mean work has stood still. It means continuity is stronger than the headlines suggest, and that the economy often repurposes old structures instead of replacing them outright.
The old debate is familiar. Smith saw new jobs emerging from specialization and the division of labor. Schumpeter saw technology destroying existing work and creating new work in waves. The Swedish evidence does not erase Schumpeter; it reorders the hierarchy. The biggest engine of job creation in the data is not the shiny machine or the breakthrough platform. It is specialization. The study separates Schumpeterian occupations, tied directly to new technology, from Smithian occupations, which arise from deeper division of labor. Most new work is Smithian, and that matters because it reveals where labor actually finds shelter.
Investors are taught to worship novelty, but novelty is often a poor guide to resilience. A company, a sector, even an entire economy can look dynamic while depending on a surprisingly old set of functions. That is the hidden fragility: when people mistake outward change for structural change. The Swedish paper suggests that most employment is not replaced wholesale by new technology; it is rearranged around older economic purposes. This is closer to a cathedral repaired over centuries than a factory razed and rebuilt in a night.
The paper’s distinction between Schumpeterian and Smithian occupations is more than academic taxonomy. It exposes a difference between what is loud and what is durable. Schumpeterian work is small and unstable, and its workforce is getting older. Smithian work, by contrast, commands the largest earnings premium. That should unsettle easy futurism. The highest rewards may not go to the most glamorous job categories, but to the ones that make the rest of the system run more smoothly through narrower specialization. The economy often pays for lubrication, not spectacle.
There is a reason this feels counterintuitive. Human beings are drawn to the visible event, not the invisible structure. We remember the steam engine, not the thousands of small occupational changes that followed it. We watch the wave, not the tide. Yet labor markets, like river deltas, are formed by sediment. A single dramatic force can be less important than the slow deposition of tiny shifts. History repeatedly shows that systems survive not by becoming pure and new, but by accreting layers of adaptation around older functions.
The study also shows that between 1990 and 2019, strongly Smithian jobs accounted for about 60% of employment, while roughly one-third was in predominantly Schumpeterian jobs, according to the syndicated column’s reporting. That split matters even if you strip away the labels. It suggests that even in the late modern period, the labor market did not tilt entirely toward technology-born work. It still leaned heavily toward jobs made possible by specialization. In a world obsessed with disruption, the most common outcome was not replacement. It was refinement.
A separate US study, using a large language model, found that roughly one-third of new jobs created from 2011 to 2023 were technology-related. That finding fits the same pattern: technology does create work, but not all work, and not necessarily the majority. The more important question is whether new tasks are complements to old systems or substitutes for them. History suggests the economy is most resilient when technology extends the chain of specialization rather than snaps it. When it does the latter, the damage is obvious. When it does the former, the labor market keeps moving almost as if nothing happened.
This is where the psychology of investors and policymakers goes wrong. They tend to linearize change. If technology is advancing, they assume old jobs are doomed. If productivity is rising, they assume labor must be shrinking. But markets and labor systems are not straight lines; they are adaptive ecosystems. Some species vanish. Others split. Many remain, merely altered. The mistake is not in expecting change. It is in underestimating the inertia of institutions, training, regulations, customer habits, and simple human preference. The world resists clean theoretical stories.
Luis Garicano, in Messy Jobs, argues that some task groupings are loosely cohesive and highly susceptible to AI disruption. That is a useful warning, but it should be held beside the Swedish evidence, not above it. The lesson is not that AI changes nothing. It is that susceptibility is uneven. Some jobs are more like bundles tied with twine; pull one thread and the bundle comes apart. Others are more like forged joints, where replacing one element does not break the whole. If you want to know what AI can dislodge, do not ask only whether a task can be automated. Ask whether the job is a coherent organism or a loose pile of chores.
That is why sweeping claims about the death of work usually disappoint. The future does not arrive as an abstract category. It arrives through friction. Managers keep legacy systems because replacement is costly. Customers prefer familiar service because coordination is hard. Workers retrain slowly because the return on learning is uncertain. Even when a new tool is objectively better, adoption lags. Game theory helps here: each participant waits for others to move first, and the equilibrium becomes sticky. The outcome is rarely optimal. It is merely stable enough to persist.
This also explains why labor markets can appear antifragile without being invulnerable. They absorb shocks by recombining tasks. But that same flexibility can conceal accumulation of risk. If too many jobs become loosely defined, the system may look adaptable right up until a new technology starts to standardize and compress them. Then the very vagueness that once allowed survival becomes a liability. Loose task groupings are easier to reorganize, but also easier to hollow out. Optionality is valuable until the option is exercised against you.
There is a familiar temptation in markets: to think the world is always moving toward the new and away from the old. Yet the Swedish study points toward a harsher truth. Most of what is paid today was already economically legible more than a century ago. That does not mean current jobs are the same in detail. It means their core function has a lineage. Old forms survive because the economy keeps needing coordination, sorting, supervision, repair, and specialization. Technology changes the method. It does not abolish the need.
The practical implication is not nostalgia. It is humility. Forecasts that treat labor as if it were a single block invite error. Better to think in terms of layers. The top layer is where technology creates visible excitement and obvious churn. The deeper layers are where specialization accumulates and where most income is made. If you focus only on the top layer, you overestimate discontinuity. If you focus only on the base, you miss the pressure that eventually reshapes the structure. The wise stance is to assume both continuity and disruption, never one without the other.
There is a final irony. The more advanced the economy becomes, the more it may depend on invisible specialization rather than headline innovation. That is because complexity creates niches. Every new machine, platform, or software system spawns maintenance, compliance, integration, support, and oversight. These are not glamorous tasks, but they are real ones. Civilization is not a parade of inventions; it is a long negotiation between invention and coordination. The winner is usually the side that can absorb change without collapsing its working parts.
So who was right, Smith or Schumpeter? The answer may be less dramatic than either man would have preferred. Schumpeter was right that technology destroys and creates. Smith was right that division of labor is the deeper engine of employment growth. The Swedish evidence leans toward Smith because the labor market is still built from specialization, not from perpetual replacement. That should not comfort anyone who wants certainty. A system that survives by rearranging old parts can still fail if the rearrangement gets too strained. But it does mean the future is likely to arrive wearing older clothes than the prophets expect.
