For most of the modern era, the concept of specialization appeared firmly defined and understood. Split a task into its smallest parts, hand each part to someone who does only that, and the system improves. It gets faster, and faster means more productive. Nobody asked if this also meant more stable, because specialized systems were measurably outproducing the alternative, the cottage industry that came before.
1776
In 1776 Adam Smith made the case for the understanding of the time using the example of a pin factory. One worker could do every step of making a pin, drawing the wire, cutting it, sharpening it, attaching the head, and they might produce a handful of pins by the end of the day. However, ten workers, each doing one step (and only that step), could produce thousands.
Smith wasn’t describing a hypothetical. He’d seen the numbers from real pin manufacturing. The division of labor made it a different order of magnitude more productive. This understanding made specialization the obvious, unquestioned foundation of everything built afterward: factories, professions, departments, the entire social milieu of how work gets organized.
Smith’s argument was about output. It said nothing about what happens when something goes wrong.
Specialization creates strength through interdependence
1893
In 1893 Émile Durkheim published The Division of Labor in Society. Durkheim wasn’t interested in pins. He was interested in what holds a society together once everyone in it has stopped doing the same jobs and social roles.
Simple societies, where most people perform the same daily activities (farm, hunt), stay bound through what he called mechanical solidarity; their lives and roles resemble each other’s, leading to shared beliefs, and shared customs in a shared daily life.
Complex societies (where a doctor, a farmer, a steelworker, and a teacher all do entirely different things) cannot be united by the sameness of their lives. Nobody resembles anybody else. But Durkheim noticed that they cohere anyway, and more tightly than before. He called it organic solidarity: a bond built from difference instead of sameness. The doctor needs the farmer to eat. The farmer needs the steelworker’s tools. Nobody can meet their own needs completely, but that mutual inability is precisely what binds them.
The division of labor produces more pins, but it also produces a society that can no longer function as separate parts. By Durkheim’s accounting, that’s a feature, not a risk. Interdependence, done well, is a stronger bond than similarity ever was. A surgeon depends on the anesthesiologist, who depends on the pharmacist, who depends on the lab technician, and none of them could do each other’s job. That dependence is not a weakness in the system; it is the system.
Interdependence describes a state: how a system holds together under normal operation. It says very little about what happens when one of those parts fails, because in a fully specialized structure, a failure in one irreplaceable part won’t stay contained; it propagates. Everyone downstream was depending on exactly that part and nothing else.
But Durkheim was describing cohesion, not resilience.
The twentieth century asked a deeper question
1955
Years later, Robert MacArthur was watching systems fail and recover in the wild, where nobody had designed the division of labor and nobody could redesign it after a bad season. In 1955, he reported that ecosystems with more species were more stable. The intuition was simple and, on the surface, close to Durkheim’s: more parts, more connections, more mutual support, therefore a sturdier whole. More differentiation, more strength, in nature just as in factories and societies.
While ecologists were working this out in the field, a British cybernetician named W. Ross Ashby arrived at a strikingly similar conclusion: control theory. In 1956, Ashby formalized the law of requisite variety, commonly summarized as “only variety can absorb variety.” A regulator (whether it’s a thermostat, a brain, or an organization) can only successfully manage a system if it has at least as much internal variety as the disturbances it needs to handle. A control system with fewer responses than the situations it faces will, sooner or later, meet a situation it has no response to at all.
1972
In 1972, a physicist-turned-ecologist named Robert May built a mathematical model of ecosystems and got a surprising result. Randomly connected complex systems become less stable as they grow more diverse, not more. Add species, add connections, and past a certain point the whole system becomes statistically unlikely to survive at all. This became known as the diversity-stability paradox. It remained a genuine, unresolved paradox for decades, because it contradicted what field ecologists could see with their own eyes: real rainforests and real coral reefs, some of the most species-dense environments on the planet, were also some of the most enduring. May’s math wasn’t wrong; it was incomplete. Figuring out what it was missing took the next thirty years.
Later researchers determined that two different things had been conflated: how many different roles a system has, and how differently the things filling similar roles respond to adaptive pressure. May’s model assumed both random connections, and identical behaviors under stress. But real ecosystems don’t work that way. Once ecologists looked specifically at how individual species behaved when conditions soured, the paradox began to dissolve.
1984
Charles Perrow, studying industrial accidents, argued in 1984 that systems combining high complexity with high levels of interconnectedness become more prone to catastrophic failure as they grow more intricate, not less, because there’s no slack left anywhere for a small problem to stay small before it cascades. He called it normal accident theory.
Looking beyond a number
1999
Shigeo Yachi and Michel Loreau published the paper that reframed the puzzle in 1999 as the insurance hypothesis. Their argument was that biodiversity doesn’t stabilize an ecosystem by adding more roles to fill. It stabilizes an ecosystem by adding more ways for the same role to endure changes. If ten different species can all perform the same function (e.g. pollination or nitrogen fixation), and each of those ten species has a different tolerance for drought, heat, cold, or disease, then no single environmental shock can take out the function itself. Something in that group of ten will keep doing the job even when nine of them can’t. MacArthur was counting species. Yachi and Loreau were asking what those species were insuring against, and the answer was each other’s temporary failure, not each other’s absence.
In 1956, Ashby wasn’t discussing pollinators or pin factories. But his law describes the same gap the insurance hypothesis filled: variety isn’t valuable because it divides labor more finely; it’s valuable because it gives a system more than one way to answer the same disturbance.
In 1984, Perrow’s normal accident theory was to high-reliability theory, what May’s paradox was to the insurance hypothesis: a caution against the assumption that more complexity, or more redundancy, is inherently safer. Redundancy without slack, layered onto a tightly coupled system, can make a failure spread faster, not slower. The lesson from both pairs is the same lesson stated twice in two different centuries and two different disciplines: it was never diversity alone doing the work. It was diversity of response, kept loose enough to absorb a shock instead of transmitting it.
2003
Four years later, a group led by Thomas Elmqvist gave this idea a name: response diversity (distinguishing it from functional redundancy). Functional redundancy is when multiple species perform the same role. But if all of those species respond to the same stressors in the same way, they all fail together. Response diversity is when redundant species fail under different conditions, at different times, for different reasons.
A system with ten redundant but similarly vulnerable parts has no more resilience than a system with only one. A system with three parts, that each fail under different kinds of pressure, has genuine insurance, even though it has fewer redundancies.
The difference lies between dividing a task, and diversifying a response to the same task. Durkheim was right that specialization creates interdependence. But he had no way of seeing that interdependence and insurance are different properties, and that a system can have a great deal of one while having almost none of the other.
Response diversity goes beyond crisis management
2004
Lu Hong, an economist, and Scott Page, a social scientist, published research in 2004 showing that a team of people using truly different problem-solving approaches could outperform a team where every member was individually more skilled but thought about the problems the same way. Page’s later book made the finding a memorable phrase: diversity can trump ability. This mechanism is the same one Elmqvist observed in ecosystems. A team of specialists who all approach a challenge with the same mindset, will all get stuck at the same dead end. A team with different ways of thinking has a better chance that someone’s approach clears the obstacles the others could not.
2007
Karl Weick and Kathleen Sutcliffe spent years studying why certain dangerous workplaces (nuclear plants, aircraft carriers, air traffic control towers), where a single mistake could kill, could operate for long periods with so few major accidents. They published their findings in 2007 as a theory of high-reliability organizations. What they found was:
- Deliberate redundancy and the encouragement of diverse perspectives (Reluctance to Simplify)
- More than one person positioned to catch the same class of failure, decision-making pushed down to whoever was closest to the problem rather than reserved for a single specialist up the chain (Deference to Expertise)
- A working culture that treated every near miss as information rather than an embarrassment to bury (Preoccupation with Failure)
- Real-time awareness of the front line situation (Sensitivity to Operations)
- A developed capacity to recover (Commitment to Resilience)
Ultimately, Weick and Sutcliffe found that these organizations succeed not by eliminating errors entirely, but by developing a deep capacity for resilience; allowing them to anticipate disruptions and contain small failures before they escalate into catastrophic ones. These organizations hadn’t read Yachi and Loreau. They had independently discovered the principle under operational pressure: their survival depended on having more than one way for a threat to get caught, not on each person’s task being narrower.
A lesson in failure and inevitability
2020
Corporate supply chains spent decades organized almost entirely around Smith’s original insight: find the single most efficient supplier, and use only that supplier. Anything else was less cost-effective and less efficient. This was the same unexamined assumption Durkheim’s framework had left: specialization and concentration, taken to their logical end, should produce the strongest possible system. The COVID-19 pandemic shattered that assumption.
Companies that had concentrated their sourcing in a single supplier or region floundered when that single source failed. Research published in the years since, including a study following more than fourteen hundred Chinese manufacturing firms through the pandemic, found that companies with more diversified supplier bases enjoyed more buffer capacity and posted stronger profitability (through both the disruption and the recovery) than companies that hadn’t diversified.
A single-sourced supply chain is, in Yachi and Loreau’s terms, an ecosystem with no insurance. It runs beautifully until the one thing it depends on fails, and then it too fails.
Today
Smith was correct that dividing a task multiplies output. Durkheim was correct that specialization builds a durable form of social cohesion through interdependence. But what happens to a specialized, interdependent system when even one of its dependencies fails? Biologists, cyberneticians, and organizational researchers answered.
Resilience (an outcome) doesn’t come from specialization; increased production (an output) does. Outcomes like robustness, and resilience come from how many different ways the system has to keep the same function alive when the first way fails. A diverse set of roles, a diverse set of perspectives, and a commitment to enduring.
Specialization can coexist with these kinds of outcomes, but it can also subtly undermine them.
How to find out which one your organization has
- Identify every critical function that would still perform if the current person or system failed. Not who covers vacation. What survives a permanent loss.
- For each backup you find, ask what would take it out too. A second supplier in the same city as the first, a second reviewer trained by the first, a second server in the same data center: same failure, different name. Response diversity means the backup fails under different conditions than the original, not just at a different time.
- Separate “we depend on each other” from “we’re covered if one of us fails.” Both are real properties, and a team can have a great deal of the first while having none of the second. Durkheim’s interdependence and Yachi and Loreau’s insurance are not the same question, and an audit that only asks the first one will miss the second entirely.
- Check for slack before adding redundancy. Perrow’s warning applies directly here: stacking a second and third path onto a tightly coupled system can spread a failure faster instead of containing it. Redundancy needs room to absorb a shock, not just a second copy standing next to the first.
- Look for teams that think or problem solve the same way. Page’s research suggests the fix isn’t a more skilled specialist. It’s a different approach in the room, even a less individually skilled one.
Pitfalls to watch for
- Mistaking headcount for insurance. Three people who trained the same way, at the same place, under the same manager, will likely fail the same way under the same pressure. That’s redundancy without response diversity, the exact gap Elmqvist’s research identified.
- Treating specialization as diversification. Redundancy (multiple suppliers, multiple reviewers, multiple approaches) costs more; to secure, to coordinate, to maintain. The insurance hypothesis explains why the cost is often worth paying. Cross-training people who will be overloaded in their own roles during a crisis, is not diversifying.
- Using Perrow as an argument against redundancy. His warning is narrow and easy to flatten: highly interconnected complexity without slack is dangerous. This is a case for building in slack alongside redundancy, not a case for skipping redundancy.
- Waiting for a crisis to determine which kind of system you built. Single-sourced supply chains looked identical to diversified ones for years, right up until they didn’t. The gap between interdependent and insured is often invisible until something breaks.
References and further reading
- Smith, Adam. An Inquiry into the Nature and Causes of the Wealth of Nations (1776).
- Durkheim, Emile. The Division of Labor in Society (1893).
- MacArthur, Robert H. “Fluctuations of Animal Populations and a Measure of Community Stability.” Ecology 36, no. 4 (1955): 533-536.
- May, Robert M. “Will a Large Complex System Be Stable?” Nature 238 (1972): 413-414.
- Yachi, Shigeo, and Michel Loreau. “Biodiversity and Ecosystem Productivity in a Fluctuating Environment: The Insurance Hypothesis.” Proceedings of the National Academy of Sciences 96, no. 4 (1999): 1463-1468.
- Elmqvist, Thomas, Carl Folke, Magnus Nystrom, Garry Peterson, Jan Bengtsson, Brian Walker, and Jon Norberg. “Response Diversity, Ecosystem Change, and Resilience.” Frontiers in Ecology and the Environment 1, no. 9 (2003): 488-494.
- Ashby, W. Ross. An Introduction to Cybernetics (1956).
- Weick, Karl E., and Kathleen M. Sutcliffe. Managing the Unexpected: Resilient Performance in an Age of Uncertainty (2007).
- Perrow, Charles. Normal Accidents: Living with High-Risk Technologies (1984).
- Hong, Lu, and Scott E. Page. “Groups of Diverse Problem Solvers Can Outperform Groups of High-Ability Problem Solvers.” Proceedings of the National Academy of Sciences 101, no. 46 (2004): 16385-16389.
- Page, Scott E. The Difference: How the Power of Diversity Creates Better Groups, Firms, Schools, and Societies (2007).