What is automation in software work?

Higher Seminar in Media and Communication Studies with James Steinhoff from University College Dublin

Automation is a central concept for political economy. The dynamic of competition means that automation is an “objective necessity imposed by the very functioning of the capitalist mode of production itself” (Ramtin 1991, 101). Most research on automation has occurred in the context of industrial production, and to a lesser degree in the service industry and gig economy, under the guise of algorithmic management. Recently, research on the generative AI-driven automation of various forms of media work, including journalism, has flourished. However, there is very little research on automation in the work of software production. Some older research even contends that software production inherently cannot be automated (Andrews et al. 2005). Generative AI now renders such conclusions dubious, as coding is one area where it is widely used already. In any case, software is a qualitatively distinct product, enabled, as it is, by the metamedium of the digital computer. In terms of automation, cloud compute substitutes for heavy machinery. This technical change has political dimensions as it entails that automation is not necessarily a process implemented by capital, but by software workers themselves.

To investigate this situation, I have conducted interviews with three categories of software-producing labour: software developers, data scientists and technical artists. Software developers are the backbone of the tech industry and an increasingly “blue collar” category of software worker which builds applications. Data scientists are perhaps the most elite category of software worker as they can create machine learning models. Technical artists combine the artistic work of a digital artist with the programming capacities of a software developer. They use this skill set to automate digital artwork, producing 3D assets at scales unimaginable by conventional means. This role appeared first in VFX and games but is now involved in AI production as 3D simulations are employed to generate synthetic AI training data.

I discuss these three kinds of work and how automation figures in them, highlighting four distinguishing aspects of automation in software work. Automation in software work is: a) accretionary, meaning that it evolves through increasing layers of mediatization, b) recursive, insofar as it recruits software in the production of software, and c) mundane, making it increasingly difficult to distinguish from labour, both concretely and analytically, insofar as the process of automating becomes the goal of labour, not something applied to labour. While the processes of automation I discuss here include the use of generative AI, they are not limited to it. Generative AI was preceded by many less exotic forms of automation in the software work context.

James Steinhoff is an assistant professor in the School of Information and Communication Studies at University College Dublin. His research focuses on the political economy of AI and other algorithmic technologies. He is interested in software work, automation, the tech industry and synthetic data. He is author of Automation and Autonomy: Labour, Capital and Machines in the AI Industry (Palgrave 2021) and co-author of Inhuman Power: Artificial Intelligence and the Future of Capitalism (Pluto 2019). He is currently working on a book about synthetic data.