Automation in O‑Ring Production: Complementary Task Dynamics and Implications for Labor Displacement

In the context of production technologies where multiple tasks exhibit quality‑complementarity—essentially an “o‑ring” dynamic—the standard assumption that tasks can be substituted one‑by‑one with automated equivalents does not hold. This NBER working paper (W‑34639, DOI 10.3386/w34639, Issue January 2026) investigates the consequences of such complementarities for automation strategy, worker time allocation, and labor income. **1. Incomplete Substitution Logic** When a worker faces a fixed amount of time across numerous tasks, replacing one task with a machine does not merely reduce the time required for that single activity; it alters the marginal utility of automating the remaining tasks. In an “o‑ring” setting, the overall output is the product of task qualities. Thus, the benefit of automating task A depends on the quality of all other tasks, and the decision to automate task B cannot be evaluated in isolation. The conventional task‑by‑task substitution model therefore underestimates the interdependence among tasks. **2. Discrete, Bundled Automation Decisions** Because the returns to automating one task influence the returns to others, firms often face discontinuous decision boundaries. Even if the quality of a machine’s output improves smoothly, the optimal adoption strategy may switch abruptly from partial to full automation of a bundle of complementary tasks. This bundling effect makes the space of feasible automation strategies highly non‑convex and encourages strategic coordination in the deployment of new technologies. **3. Labor Income May Rise Under Partial Automation** Contrary to the intuitive narrative that automation simply supplants labor, the study shows that partial automation can actually increase worker earnings. By automating lower‑quality tasks, a worker can reallocate time toward the remaining bottleneck tasks whose output drives the entire production function. The marginal product of time dedicated to these critical tasks rises, so the worker’s wage—linked to the value of their contribution—can increase. This result challenges the prevailing view that automation is inherently displacing. **Implications for Exposure Indices** Current exposure metrics compute an aggregate risk of displacement through a linear aggregation of task‑level automation probabilities. The findings in this paper indicate that such indices are ill‑suited when tasks are quality complements. What matters is the network of bottlenecks and how automation reshapes the worker’s allocation of time around those critical tasks, rather than merely the average exposure. Consequently, policymakers and researchers should refine exposure assessment tools to incorporate task interdependencies and the non‑linear effects of partial automation. **Conclusion** The paper underscores that automation strategies in complementary‑task environments must account for complex substitution dynamics, discrete decision boundaries, and the counterintuitive possibility of income gains for workers. These insights invite a reevaluation of prevailing models of labor displacement—and highlight the need for more nuanced, structure‑aware analytic frameworks in the era of rapid automation.