Technology and Organizations Shape Each Other. That Is Not a Metaphor.

When organizations adopted email in the 1990s, they did not just add a new communication tool to their existing workflows. Email changed how decisions got made, flattened some hierarchies while creating new bottlenecks, and gradually redefined what responsiveness meant in professional settings. The change ran in both directions. Email shaped organizations, and organizations shaped how email was used. New norms emerged. New problems emerged. New organizational roles emerged. Eventually, new tools emerged to fix the problems email created, and those tools then changed things again.

That recursive back-and-forth is what I mean by coevolution. Technology and organizations do not evolve on separate tracks that occasionally intersect. They shape each other over time in a process that is ongoing, recursive, and not fully predictable in advance.

Orlikowski (1992) established the theoretical foundation for this in her structuration-based view of technology. She argued that technology is both a product of human action, designed and built through social and organizational processes, and a medium of human action, shaping what people can do and how work gets done. The relationship between technology and organization is therefore not linear. Technology is produced by organizational actors, deployed into an organizational context, appropriated through patterns of use, and through that use, becomes part of the structure that shapes future action and future technology choices. This is what she called the duality of technology, and it is the mechanism underneath what I am calling coevolution.

Leonardi (2011) extended this by introducing the concept of imbrication. Human and material agency interlock recursively, like the overlapping scales on a fish. Each round of use produces new routines and new material arrangements that become inputs to the next round. The technology does not determine outcomes and people do not simply choose outcomes independently of the technology. Outcomes emerge from the interlocking of both over time. Leonardi gives co-evolution a more concrete mechanism than structuration alone provides, showing how specific material and human actions layer on top of each other across episodes.

Benbya, Nan, Tanriverdi, and Youngjin (2020) connect this to complexity science. Their argument, which is verified in my study materials, is that digital systems exhibit emergence, coevolution, and chaos. Emergence means that system-level behavior cannot be predicted from the properties of the components alone. Coevolution means that technology, organizations, and environments adapt to each other continuously, not in a single adoption event. Chaos means sensitivity to initial conditions: small differences in how a technology is implemented or used early can produce large divergences in outcomes over time. The implication is that linear models of IT adoption and impact, the kind where you measure use at time one and performance at time two, miss the recursive adaptive process that actually produces outcomes.

The historical arc of IT infrastructure makes this visible if you look at it over enough time. Mainframe computing in the mid-twentieth century created centralized IT departments as a new organizational function, which then shaped what technology was built to serve. Client-server computing in the 1980s and 1990s enabled distributed processing, which enabled the decentralization of business units, which created demand for new integration tools because the decentralization produced coordination problems. The internet and then cloud computing enabled further distribution and eventually created the conditions for globally distributed teams, which then created demand for collaboration platforms, which then created new coordination structures, which are now generating demand for AI tools that manage the complexity those structures created.

Each transition is not just a technology story. It is an organizational story. New roles appeared: the DBA, the systems administrator, the cloud architect, the DevOps engineer. Old roles disappeared or transformed. Business processes were redesigned around technical constraints and then redesigned again as those constraints shifted. The technology and the organization moved together, each shaping the other's trajectory.

What makes the current moment with generative AI interesting from a coevolution perspective is how fast the feedback loops are running. Previous infrastructure transitions played out over decades. The shift from on-premise to cloud took the better part of fifteen years to become the dominant mode for most enterprise applications. Generative AI capabilities became widely accessible in 2022 and 2023, and organizations are already redesigning processes, restructuring teams, and rethinking hiring in response. The coevolution cycle is compressed.

But compressed does not mean different in kind. The same recursive logic applies. Organizations that have adopted AI writing assistance have started changing what human review processes look like. Changed review processes create new requirements for AI systems (the tools need to support tracked edits, maintain version history, flag uncertainty). Those requirements shape the next generation of tools. The technology shapes the organization, which shapes the technology, which shapes the organization again. That is not metaphorical. It is the mechanism.

The organizational difficulty is that coevolution makes prediction genuinely hard. You cannot evaluate a technology adoption decision only in terms of what the technology does today. You have to think about how adopting it will change your organization, and then how that changed organization will interact with the next version of the technology, and the one after that. Most technology evaluation frameworks do not have a good way to model that. They are point-in-time assessments of a relationship that is, by its nature, time-extended and recursive.

This is also why structuration theory helps explain why the same tool produces different outcomes across organizations. The coevolution path depends on the starting conditions. An organization that brings in AI tools with strong existing processes for evaluating information quality will coevolve differently with those tools than one that brings in the same tools with weaker quality standards. The technology is identical. The starting structural context is different. The coevolutionary path diverges.

The uncomfortable implication is that there is no stable endpoint. The relationship between technology and organization is not moving toward an equilibrium where the right organizational form for AI has been discovered and can be copied. It is a continuing process of mutual adaptation, and the organizations that do well are the ones that can adapt their structures and practices as the technology changes, rather than treating a single implementation moment as the destination.