Both OSI and TCP/IP are packet switching models, but for better or worse, the sevenlayer OSI model that Bachmann described never really took off. The simplified TCP/IP four-layer model described by Cerf ’ s group did get traction, and based on those early adoptions, the Internet as we know began in earnest. <sup>31</sup> TCP/IP was ratified in 1980, and besides compressing the stack geometry by combining the top three layers of the OSI stack into one, it proved a much simpler and flexible solution for early industrial and government network adopters. <sup>32</sup> TCP/IP “ won ” not simply because its early adopters were more visionary but because it just worked better to link together heterogeneous existing systems and translating between them so that they could work as one. America ’ s factories were not linked by the organicist tissue of TRON; they were full of various proprietary computing systems running mutually incompatible software. TCP/IP made it much easier to design and implement general-purpose hardware and software that could link all these together — as is — and thereby quickly realize gains in efficiencies with the machines that managers already had. As more sites (and more nodes and more Users) connected systems through this network platform (factories connected to suppliers to headquarters to warehouses and so on), the network became that much more valuable to every connected User. As the platform that glued all these together became more established, the opportunity cost of using alternatives such as OSI increased (a good example of generative entrenchment). We see that platforms that allow for a tactical appropriation to optimize existing systems have an advantage over those that would appear from a tabula rasa, at least within this context for established industrial systems. Over time, the entirety of those original economies and mechanisms may be replaced bit by bit by new technologies and new communicative norms that are better tuned to take advantage of the wider systems that emerge.
Platform and Stack, Model and Machine
63