The Jack-of-All-Trades Wins Alone. The Crowd Loses Together.

Based on the research ofYang, Wang and Johnson, "A Generalist Activity Paradox: Individual Performance and Collective Production Efficiency on Open Innovation Platforms," MIS Quarterly, 2026

Yang, Wang, and Johnson built their study on the simple fact that open innovation platforms for new product development do not hand contributors one job. They decompose a product challenge into many task types, idea generation, refinement, evaluation, combination, documentation, and let contributors move between them however they like. Call that movement generalist activity: how widely a contributor spreads their effort across the available task types rather than sticking to one. Drawing on the generalist literature and the theory of collective production, the authors build a multilevel framework that tracks the same behavior at two levels at once: what it does for the person doing it, and what it does for the crowd they are embedded in.

The two answers point in opposite directions. At the individual level, generalist activity is rewarded: contributors who range across task types perform better than contributors who stay narrow. At the crowd level, the opposite empirical pattern holds. A crowd in which generalist activity is more prevalent produces less efficiently as a collective. The tension is not a wash, it is a structural paradox, because the same decomposition that makes roaming possible is also what makes a crowd of roamers worse at finishing the product. The one piece of good news in the finding is that the tension is attenuated as the crowd grows larger, meaning scale is a partial, not a complete, cure.

Why the Crowd Can't Just Reward Its Best All-Rounders

The paradox would be a curiosity if it stopped at correlation. It does not. The authors' mechanism analysis gets at why the crowd-level effect is negative, and the answer reframes what "valuable contribution" even means in these settings. Valuable contributions are more likely to emerge when a contributor with generalist activity builds on top of task-focused input that an earlier, low-generalist-activity contributor already laid down. In plain terms: the breakthrough idea tends to be a generalist's recombination of something a specialist nailed down first, not a generalist's recombination of another generalist's half-finished sketch.

That is the piece every operator tends to miss when they look only at individual leaderboards. A platform that rewards roaming on the individual dashboard is implicitly asking more contributors to roam. But if nobody is laying the focused, narrow, "boring" groundwork a recombination needs, there is less of it to recombine. The crowd's upstream supply of task-focused inputs dries up precisely because the behavior that wins individually is the behavior that depletes it. This is the theory of collective production made concrete: a crowd's output is not the sum of its members' individual bests, it is a function of how their contributions chain together, and that chain has a load-bearing link, the specialist's narrow input, that nobody is individually incentivized to supply.

InnoCentive, Topcoder, and Kaggle Already Live This Tension

None of this is unique to one platform. It is close to the operating condition of every crowdsourced innovation marketplace that splits work into many task types. InnoCentive's challenge model, now run under Wazoku, structures work around "Seekers" who post a problem and "Solvers" who submit a response; a Solver can engage with physical-science, life-science, or business-model challenges across the catalog rather than specializing in one problem class, which is exactly the kind of task-type breadth the paradox describes. Topcoder runs the same logic inside software and data work: contributors can move across development, design, and data-science challenge tracks on one account, and academic work on crowdsourced software development has specifically studied how Topcoder clients decompose a project into smaller tasks before releasing them to the crowd. Kaggle, which now frames itself as a crowdsourced AI proving ground running competitions, benchmarks, and hackathons side by side, gives every registered user the same open invitation to jump between an image classification contest this month and a tabular regression contest the next.

The research record backing the individual half of the paradox is not limited to this one paper, either. Hwang, Singh, and Argote's 2019 study in Information Systems Research, "Jack of All, Master of Some," found that contributors who helped across a broad range of topics in a customer-support crowdsourcing community were more likely to generate novel product ideas later than contributors who had stayed narrow, breadth built the associative capacity that produced novelty, an individual-level payoff for exactly the kind of range-roaming the Yang, Wang, and Johnson paper calls generalist activity. On the organizational side, Fahrenkopf, Guo, and Argote's 2020 Organization Science study of personnel mobility found that new members with specialist backgrounds transferred their knowledge into a receiving group less successfully than generalists did, especially when the receiving group itself was organized generally rather than by specialized roles, a structural echo of the same tension: specialization is an asset to the system that is hardest to make portable or individually rewarding.

Put together, three separate literatures are converging on the same point from different angles: generalism keeps winning the individual contest, and the collective keeps needing the narrow work that individual contest does not reward.

The Fix Is Protecting the Groundwork, Not Banning the Roamers

A crowd made entirely of people chasing the next task type is a crowd with nobody left minding the task it is on.

The operational takeaway is not to punish contributors for ranging widely. The paradox's own individual-level finding says that roaming is genuinely productive for the person doing it, and platforms that tried to flatten that reward would just suppress genuine individual performance along with it. The takeaway is to stop treating the leaderboard as a proxy for what the platform needs in aggregate. A leaderboard built purely on individual output will, by construction, keep pulling the crowd's composition toward more generalists, and the paradox's mechanism analysis says that composition shift quietly starves the pipeline of focused inputs the best recombinations depend on.

The crowd-size finding gives operators a second, less obvious lever: the tension is not fixed, it shrinks as the crowd grows. A small challenge community cannot afford to lose its specialists to the pull of individual reward; a much larger one has enough contributors that a meaningful specialist population survives even while most contributors drift toward range. That argues for being most careful about leaderboard design and task-matching exactly when a platform or a specific challenge is small and growing, and for treating scale itself as part of the fix rather than assuming good individual incentives alone will keep the collective engine running.

Before you redesign a reputation system to reward breadth, or celebrate the contributor who shows up in every challenge category, ask what happens to the pipeline of narrow, unglamorous, task-focused work your best recombinations actually depend on. That work does not show up near the top of an individual leaderboard. On these platforms, it is apparently the reason anything near the top of that leaderboard is possible at all.