Research

Abstract: Centralization can improve operational efficiency by coordinating interdependent activities, yet impede adaptation to disruptions when changing local conditions must be communicated and interpreted at the center. When a centralized AI system performs these activities, its information-processing capabilities may ease these demands. How does this affect adaptation? I study one of the world’s largest container terminals, where the same AI dispatches autonomous trucks through zones used almost exclusively by autonomous trucks (pure-AI) or shared with human-driven trucks (human-AI). Across 179,925 tasks, pure-AI zones move 9.8% more cargo per vehicle during ordinary operations. During recurrent mechanical faults, pure-AI zones recover more slowly, equalizing per-vehicle productivity between zones. During unprecedented geomagnetic storms, pure-AI zones lift 12% less cargo than human-AI zones. A formal model shows how a configuration favored under a historical distribution of disruptions can lose its performance advantage when that distribution changes. The findings show that when an AI system performs organizational activities, the capacity to adapt depends on whether disruptions resemble conditions the system has encountered before.

Abstract: Research on GenAI and work has concentrated at the task level. We ask when task-level effects alter knowledge-work industry architecture: the allocation of activities across firm boundaries. Using a three-stage abductive design—an executive roundtable, a panel of 626 firm websites (2017–2025), and 33 interviews plus two further roundtables with C-level executives and corporate directors—we document divergent repositioning: professional services firms foreground relational value, while interface-oriented technology firms emphasize scalable interfaces. We find that activity reallocation is shaped by two independently varying dimensions: expertise bypass and exchange governability. Their intersection yields four configurations: contestable reallocation, persistent bundles with selective insourcing, federated expert provision, and integrated expert provision. We extend industry architecture theory by explaining change when production and exchange conditions shift separately.

Abstract: How do senior decision-makers assess the competitive implications of GenAI? Using an abductive, multi-stage study of 531 UK corporate directors (exploratory roundtables, survey, matched stimulus roundtables, and follow-up interviews, 2024–2026), we document that directors distinguish three dimensions of GenAI’s competitive implications: competitive differentiation, firm displacement, and industry disruption. Expectations vary markedly even within the same sub-sector. This heterogeneity is patterned by how directors frame their firms’ value propositions—especially perceived modularity and reliance on pattern recognition—and by complements and buffers such as proprietary data, relationship-intensive delivery, and regulatory/liability constraints. Across waves, directors’ competitive expectations remain stable, consistent with durable interpretive frames during early diffusion.

Abstract: Across industries, AI is taking over tasks once performed by humans, yet humans are often kept “in the loop” when systems reach exceptions. We study human intervention at one of the world’s largest ports in China, where autonomous trucks can be remotely taken over when stuck. Exploiting a capacity constraint and matching on pre-treatment operational characteristics, we find that human takeover reduces task completion time by roughly 7%, with gains concentrated when exceptions are neither too simple nor too complex. However, humans do not add value in selecting which trucks to take over, partly due to an attention bias. Our results align with Garicano’s (2000) management-by-exception logic but point to an important extension: in dynamic AI systems, which exceptions should be handled by humans may not be immediately obvious, underscoring the importance of carefully designing human-in-the-loop rules.

Abstract: We examine how digital infrastructure transforms coordination mechanisms and synergy patterns within multi-business firms. Through a longitudinal study of a large retail conglomerate's evolution from traditional M-form to internal ecosystem, we document a three-stage process whereby digital capabilities progressively alter value creation dynamics. Our findings reveal how digital infrastructure enables systematic lateral coordination between previously unrelated businesses, replacing sporadic collaboration with data-driven discovery of synergistic opportunities. We identify novel forms of value creation—superadditive demand synergies and data-network effects—that differ from traditional resource-based relatedness. The corporate center's role evolves from orchestrating synergies to maintaining digital infrastructure while enabling autonomous collaboration. These findings explain how digitally enabled firms successfully pursue broader, seemingly unrelated scope through fundamentally different governance mechanisms than traditional conglomerates.