Humans and Technology: Forms of Conjoined Agency in Organizations

Source: Alex Murray, Jen Rhymer, David G. Sirmon; Academy of Management Review 46(3):552-571 · 2021

ABSTRACT Organizations are increasingly deploying technologies that have the ability to parse through large amounts of data, acquire skills and knowledge, and operate autonomously. These technologies diverge from prior technologies in their capacity to exercise intentionality over protocol development and/or action selection in the practice of organizational routines, thereby affecting organizations in new and distinct ways. In this article, we categorize four forms of conjoined agency between humans and technologies: conjoined agency with assisting technologies, conjoined agency with arresting technologies, conjoined agency with augmenting technologies, and conjoined agency with automating technologies. We then theorize on the different ways in which these forms of conjoined agency impact a routine’s change at a particular moment in time as well as a routine’s responsiveness to feedback over time. In doing so, we elaborate on how organizations may evolve in varied and diverse ways based on the form(s) of conjoined agency they deploy in their organizational design choices.

Introduction. “Change in organizational routines is intrinsic to organizational routines so long as human agents perform them.”

(Feldman, 2000: 627) Routines— repeated patterns of behavior—are critical to our understanding of organizations (Cyert & March, 1963; Nelson & Winter, 1982; Parmigiani & Howard-Grenville, 2011).

Routines offer organizations both a stabilizing force as well as an apparatus to evolve with changing environmental demands (Feldman, 2000; Howard-Grenville, 2005; Pentland, Hærem, & Hillison, 2011). As a stabilizing force, routines enable organizations to consistently and repeatedly accomplish tasks, even those that are highly complex (Ashforth & Fried, 1988; Cohen & Bacdayan, 1994; Gersick & Hackman, 1990). As an apparatus for evolution, routines provide flexibility to adapt to new and changing demands (Feldman, 2000; Feldman & Pentland, 2003; Pentland & Reuter, 1994; Turner & Rindova, 2012). Recognizing the apparent tension in routines serving as both a source of stability and flexibility, recent work has focused on how routines “are produced and reproduced, and to what extent the patterns remain stable versus change over time” (Parmigiani & Howard-Grenville, 2011: 421).

Central to this line of inquiry have been studies that dive into the “black box” of routines where humans and nonhumans—technologies that include tools and artifacts—interact in an “ensemble” (Pentland et al., 2011). While humans and nonhumans together can jointly produce gains for organizations, the driving force behind a routine’s stability or evolution has rested with humans as they alone possessed a temporally-embedded capacity to act with intent (c.f., Emirbayer & Mische, 1998; Giddens, 1984; Howard-Grenville, 2005). However, in recent years, technologies in a variety of organizational settings have rapidly advanced in their capacity to parse through large amounts of data, acquire skills and knowledge, and operate autonomously (e.g., Agrawal, Gans, & Goldfarb, 2018; Beane, 2019; Faraj, Pachidi, & Sayegh, 2018; von Krogh, 2018). For instance, blockchain-enabled smart contracts are automatically distributing payments to firms’ exchange partners based on the achievement of key milestones in complex supply chains (Werbach, 2018), structured machine learning algorithms are identifying and recommending candidates in firms’ hiring processes (Greenfield & Griffin, 2018), and smart robots equipped with artificial intelligence are making autonomous care decisions for the elderly (Sanyal, 2018). Importantly, these technologies differ from other advanced technologies in their capacity to make determinations by themselves, as well as evolve their determinations over time once they are deployed in an organization.

When embedded in human-nonhuman ensembles, these technologies fundamentally alter our understanding of how and why routines evolve or remain stable. Specifically, these technologies affect routines in distinct ways based on their differing capacities to develop protocols, determine rules and guidelines for what to do (Faraj & Xiao, 2006); and/or select actions, make choices of what to do (Feldman, 2000; Howard-Grenville, 2005)—both of which are foundational components of organizational routines. We describe these emerging types of technology as agentic since they themselves possess a temporally-embedded capacity to intentionally constrain, complement, and/or substitute for humans in the practice of routines. This shifts the locus of agency in protocol development and/or action selection away from the exclusive dominion of humans; thus affecting our understanding of how humans and nonhumans interact to provide either a stabilizing force or an apparatus for evolution in new ways. Yet, despite the increased prevalence of agentic technologies in organizational work, as scholars, we understand little about how this shift in the locus of agency affects the practice of routines.

We theorize that when various types of technology, including those with agentic properties, are embedded in the human-nonhuman ensemble, different forms of conjoined agency, which we define as a shared capacity between humans and nonhumans to exercise intentionality, emerge. Thus, in this article, we focus our theorizing on the ensemble between humans and technologies to investigate the following question: How do different forms of conjoined agency impact organizational routines? To address this question, we first separate routines into the distinct components of protocol development and action selection; describing why both are meaningful to a functioning routine.

Related work. Technology in Routines Technologies—nonhuman tools and artifacts—are fundamental in the practice of routines (e.g., Pentland et al., 2011). Organizations’ rationale for developing and adopting many technologies has been to increase humans’ efficiency and/or effectiveness in the practice of routines. For example, tractors enable farmers to aerate, fertilize, and seed more efficiently.

Computer-aided design (CAD) software not only increases architects’ efficiencies in designing buildings, but also their effectiveness in conceptualizing highly complex forms. When embedded in the human-nonhuman ensemble, these technologies generate organizational gains by abating many human foibles such as inconsistency, laxity, fatigue, and unreliability; while still allowing humans to exercise intentionality over a routine’s practice. Indeed, Kaplan (2011), in her study of an organization’s use of PowerPoint software, revealed how the technology afforded humans possibilities for displaying abstract ideas through slides that could be moved and cut without affecting other slides. Similarly, Edmondson and colleagues (2001), in their study of a cardiac surgery technology’s adoption at several hospitals, found that cardiac procedures became faster and more consistent since the new technology shaped how humans performed the surgery routine even though humans continued to drive the procedure’s practice.

Despite these technologies’ level of sophistication, and the associated gains reaped by organizations that use them, they fall short of fully mitigating the human shortcomings of myopic search—humans’ predisposition to look nearby rather than afar for problem solutions; and bounded rationality—humans’ predilection to draw on biases and heuristics to select satisfactory rather than optimal actions due to limits in time, information availability, and capacity to process information (Cyert & March, 1963; Gavetti, Greve, Levinthal, & Ocasio, 2012; March, 1978; March & Simon, 1958; Simon, 1947, 1955). As such, emerging technologies are being developed to address these longstanding human shortcomings and, indeed, begin to do so through their endowment of agentic properties that enable them to intentionally constrain, complement, and substitute for humans in the practice of routines. As these technologies become more prevalent in organizational work, understanding how they differentially impact the degree and predictability of a routine’s change as well as a routine’s responsiveness is increasingly important for organizational scholars. We begin this effort by categorizing four forms of conjoined agency present in contemporary organizations.

Method. FORMS OF CONJOINED AGENCY Conjoined agency constitutes a shared capacity between humans and nonhumans to exercise intentionality. Humans have long worked alongside technologies—tools and artifacts—to practice organizational routines. Yet, agentic technologies—those that possess the capacity to intentionally constrain, complement, and/or substitute for humans in a routine’s practice—shift the locus of agency away from humans in protocol development and/or action selection. This results in new and distinct forms of conjoined agency. We therefore categorize four distinct forms of conjoined agency based on which actor—human or technology—has the capacity to exercise intentionality over protocol development and/or action selection (see Figure 1).

Conjoined Agency with Assisting Technologies As illustrated in the top left quadrant of Figure 1, conjoined agency with assisting technologies exists when the technology in the human-nonhuman ensemble (1) does not have the ability to develop protocols, and (2) does not have the ability to select actions. Numerous studies have addressed this form of conjoined agency wherein assisting technologies, which encompass an array of tools and artifacts such as cardiac surgery machines (Edmondson et al., 2001), virtual collaboration tools (D’Adderio, 2001, 2003), PowerPoint software (Kaplan, 2011), and Excel spreadsheets (Spee et al., 2016), are wielded by humans (e.g., Anthony, 2018; Beane, 2019; D’Adderio, 2010). Specifically, such studies have emphasized how humans use (or avoid using) these technologies in their practice of routines (Leonardi, 2011).

On one hand, introducing new assisting technologies in organizations has been shown to provide humans with previously unavailable information (Leonardi, 2007), alter advice networks across subgroups (Leonardi, 2013), and facilitate organizational change when new routines emerge from humans’ use of the technology (DeSanctis & Poole, 1994). On the other hand, it has been shown that humans often attempt to circumvent the implementation of assisting technologies (Leonardi, 2011) by retrofitting new technologies to maintain existing routines (Tyre & Orlikowski, 1994), using new technologies in unintended ways (Boudreau & Robey, 2005), and resisting the use of new technologies altogether (Lapointe & Rivard, 2005). Such efforts often stem from the impending limitations that assisting technologies place on the ways a routine can be practiced. Illustrating this, Boudreau and Robey (2005), in their study of an enterprise resource planning system’s introduction at a large government agency, show that individuals first resisted using the technology and then found ways to work around the constraints it introduced by using the technology in unintended ways. Taken together, the impact of assisting technologies on routines is highly dependent on the humans who wield them (e.g., Boe-Lillegraven, 2019; Kho, Spee, & Gillespie, 2019; Pentland & Feldman, 2008; Volkoff, Strong, & Elmes, 2007).

To illustrate the use of an assisting technology in the practice of an organizational routine, we draw upon The New Yorker—a weekly periodical with a circulation of over one million magazines. One critical routine practiced by The New Yorker’s editorial team is selecting the cover story for each issue. In the practice of selecting an appropriate cover story, the editorial team considers many variables including topic trends, story salience, and author-specific criteria such as reputation and quality. The use of evaluation software, an assisting technology, is critical in the practice of this routine since it allows the editorial team to efficiently sort, analyze, and identify articles based on the organization’s objective of maximizing engagement potential with readers. In this way, the evaluation software is a tool wielded by the editorial team who apply procedures and rules (i.e., protocols) to rank articles and determine which one will be elevated to the magazine’s cover story (i.e., select an action).

Discussion. To understand how emerging technologies are impacting contemporary organizations and their routines, it is necessary to explore the human-nonhuman ensemble in greater depth. In doing so, we categorize four forms of conjoined agency in which humans and technologies differentially share the capacity to exercise intentionality over protocol development and/or action selection in the practice of routines. The four forms of conjoined agency that we identify affect the degree and predictability of a routine’s change at a particular moment in time as well as its responsiveness to feedback over time in new and distinct ways. Through our theorizing, we contribute to the literatures on organizational routines and technology in organizing, and also respond to calls for theory to address the changing nature of organizations and organizing (e.g., Faraj et al., 2018; Murray et al., 2019; von Krogh, 2018).

Organizational Routines The tension of routines serving as a source of stability and an apparatus for evolution has been the impetus for much research that dives into the black box of routines (e.g., Feldman & Pentland, 2003; Pentland & Reuter, 1994). These studies have shown that routines change at a particular moment in time based on how humans enact them, and that routines respond to feedback over time as a result of humans making adjustments when suboptimal outcomes occur (Feldman, 2000; Howard-Grenville, 2005). We theorize that the distinct forms of conjoined agency present in contemporary organizations impact routine change and responsiveness in different ways. Specifically, where the locus of agency resides—with humans or technologies— in protocol development and/or action selection determines the degree and predictability of a routine’s change at a particular moment in time. This diverges from our understanding of routine change when the locus of agency for each of these components rests squarely with humans (e.g., Dittrich & Seidl, 2018). Moreover, this makes salient the importance of treating protocol development and action selection as distinct routine components.

We also theorize that the form of conjoined agency present in a routine impacts the ability of a routine to respond to feedback, and the mechanisms through which the response occurs. At one extreme, a routine can be highly unresponsive when conjoined agency is formed with arresting technologies, requiring consensus among many actors before adjustments can occur. On the other extreme, a routine can be hyper-responsive when conjoined agency is formed with automating technologies, enabling adjustments based on the technology’s real-time determinations. In the middle, a routine can have a limited level of responsiveness when conjoined agency is formed with augmenting technologies, requiring humans to contradict the technology’s suggestions. Each of these forms of conjoined agency present mechanisms that facilitate a routine’s responsiveness in unique ways when compared to conjoined agency with assisting technologies. As discussed below, this has direct implications on where different forms of conjoined agency are deployed in an organization.

Technology in Organizing Our categorization of forms of conjoined agency is also important for organizational theorists more generally as it helps direct theory development around the act of organizing in contemporary organizations. Though this requires further inquiry, it is likely that where different forms of conjoined agency are deployed in an organization will enable (or inhibit) distinct capabilities (Sirmon, Hitt, & Ireland, 2007) and competitive advantages (Sirmon & Hitt, 2009).