Introduction
Algorithms and automation are woven into metaphors of visibility and invisibility. From the Foucauldian concept of the ‘gaze’ (Foucault, 1995) to the ‘black box’ (Pasquale, 2015), the imbalance of seeing and being seen shapes an understanding of the relationship between data collection, the hidden mechanics of analytics and the enactment of control. This relationship has further been exacerbated by the complexity of datafication, algorithmic systems and AI. In recent years, with the advent of smart technologies and artificial intelligence, the visual is analysed as a placeholder for surveillance (Andrejevic, 2019), platformisation (MacKenzie & Munster, 2019) and classification (Chun, 2016). In the complex relationship between visuality and control, outside of establishing a connection between seeing and control, these technologies aim to automate the very act of seeing. Images are put into new relations of valuation and extraction determined by the affordances of large databases, platforms and algorithms, shifting the function of the image from being seen to being incorporated within operations of selective visuality, invisuality and power (Farocki, 2004; Parikka, 2023).
I interrogate the genealogy of this relationship by tracing the legacy of industrial practices of abstraction and control over labour in what I call ‘executable diagrams’: diagrams used to analyse, standardise and automate the labour process. A key example of these executable diagrams is the case of robotic process automation (RPA). RPA builds on the early industrial practice of process diagrams, which foregrounds a specific interdependence between what Melissa Gregg (2018) calls ‘temporal regimes’ of capitalism and the visual language of diagrams and process charts. However, rather than being an example of linear continuity between practices of control in early capitalism and contemporary datafied economies, the diagram acts as a hedging device that furthers processes of abstraction and translation between different infrastructural paradigms and enables the transition from human labour in the ‘temporal regimes’ of capitalism to digital data in the new regimes of data accumulation and extraction (Sadowski, 2019).
The process diagram (or chart) is an example of what Vilem Flusser (2011) calls ‘technical images’ – images that are ‘the consequence of some theory and thus mean some of its terms’ (ibid.:127). These images are technical, not necessarily in the way they are produced but, rather, in the way they operate. The relationship to representation, inherent in traditional images, is not central. Instead, they are defined by their role in enabling relations of production, extraction and control, a role evident in the use of process diagrams for visualising process abstraction in general and the abstraction of the labour process in early scientific management in particular. In this sense, they are akin to what Farocki (2004) and Parikka (2023) call ‘operative’ or ‘operational’ images but, while Farocki and Parikka reflect on the relationship between visuality, digital data and surveillance, the diagram points to a longer genealogy that incorporates analogue modes of visuality and control and underscores the ways in which labour is implicated in the operationality of the image.
This study builds on a qualitative research project carried out in 2019 at a major Australian bank. The research, conducted by an interdisciplinary team at Western Sydney University, was commissioned by the bank as internal research into the changing notion and practices of leadership in an organisation that was progressively automating many of its functions and implementing a variety of AI-powered and data-driven solutions, such as IBM Watson, intelligent chatbots and robotic process automation. I am using examples from the focus groups conducted as part of this research to talk about the knowledge politics of RPA and the relationships between workers and the work process upon which it is based. In addition, my study builds on an analysis of the functionalities of different RPA tools and the marketing materials of RPA companies.
Automation and labour abstraction
In 2019, Bank X, one of the major banks in Australia, was faced with two important developments that prompted a need to analyse its own work practices. A recent report by the Royal Commission into Misconduct in the Banking, Superannuation and Financial Services Industry (Hayne, 2019) had examined the performance of financial institutions in the country, with a focus on practices of lending and insurance and their negative impact on customers. The inquiry put forward a series of recommendations for ethical practices, stricter regulation and better standardisation in the industry, which all banks were scrambling to implement in their organisational policies while aiming to establish consistency within and across departments. The second development, one that Bank X had been contemplating for some time, was the increased automation of internal workflows in the company. In the span of the previous couple of years, the bank had rolled out a number of AI-powered and data-driven solutions in its business operations. These solutions included the AI knowledge base query system IBM Watson, introduced internally as a way for employees to quickly find the relevant policies for each case; a new intelligent chatbot trained and launched to answer customer inquiries; and RPA tools for the automation of specific tasks in different departments. Faced with the need for better regulation and the growing automation of its own processes, the bank sought consultancy outside of the usual big companies like Gartner, Boston Consultancy, PriceWaterhouse Coopers and Deloitte and contacted Western Sydney University. The multidisciplinary team led by Liam Magee included social scientists working on digital labour (myself included), a human–machine interaction specialist and a member of the administration of the university specialising in industry collaborations.
Between August and September 2019, we conducted six focus groups with management, team leaders and employees working on implementing RPA, AI-powered chatbots, overall IT solutions in the company and design strategy, as well as employees working directly with customers and using these automated technologies. We also made site visits to the company archive and customer service departments where we discussed with employees the overlaps and frictions in practices of organisational knowledge transfer and the use of automated tools. These focus groups and discussions provide an important insight into the strong interrelation between knowledge production and transfer and the automation of labour.
Knowledge of legislation, company policies and the ability to discern and make decisions on individual cases were mentioned as key skills for the workers in banking institutions, particularly workers who advise clients about their options for loans, insurance and other financial decisions. Workers who answer customers’ questions or advise them about financial products need an understanding of company policies and legislation and the ability to see how the vast information contained in these policies can be applied to individual cases. This primarily cognitive side of bank labour was one of the main areas where the company had implemented what it referred to as a technologically ‘augmented’ workplace. This has brought significant shifts in the authority and control over organisational knowledge.
For instance, while procedural and policy knowledge was something that in the past had been acquired over years of experience in the company and was a point of pride for veteran workers, the IBM-powered new query system was now offering a more immediate but also more black-boxed access to this knowledge. Workers could type a question and get the appropriate policy and procedure for a case without the need for previous experience working on similar cases. This shift in the politics of cognitive labour at the bank shows a dynamic described in the 1970s by Harry Braverman (1998) whereby the control over the production and transfer of knowledge about the labour process is abstracted and taken away from the workers. In other processes of automation, such as RPA, the dynamic of abstraction shows a more direct link between the abstraction of labour processes and the development of digital automation tools. In order to automate concrete processes, the Bank’s Robotics team had to work closely with workers whose tasks were being automated. They spent time with the team and consulted with the ‘subject matter experts’ – i.e. the employees performing these tasks. These collaborative exercises associated with automation signify the contradictory dynamics around the organisation-wide introduction of AI and automation. While the company promotes cooperation and collaboration across teams and by employees whose tasks are being automated, the sense of control over the transfer of knowledge is limited to the types of small-scale processes automated through RPA and remains extractivist in nature.
The example of Bank X also sheds light on an important aspect of automation, in which the use of visual tools becomes critical: rather than being a homogeneous process, automation occupies heterogeneous and often incongruent technological spaces. These incongruities are partially defined through the use of different proprietary digital systems with incompatible data formats (Hristova, 2020). In a much more generalised sense, however, the incongruity is also the effect of what authors call ‘infrastructural discontinuities’ (Cole, 2023) – the constellation of technological, economic and production paradigms that differentiates the techno-economic systems of capitalism and distinguishes ICT from contemporaneous datafied economies. These different infrastructural paradigms within the same company mean that there is a constant necessity to ensure technological interoperability. Even more importantly, such paradigms also create the need to negotiate how labour is incorporated and abstracted in different infrastructural relations. These relations reveal not just how labour is subjected to processes of real subsumption (Marx, 1976) through which it is standardised and optimised (a method in which the process diagram plays a pivotal role) but also how abstract labour as a relation shaped by the capitalist logic of valuation and control is further subsumed and abstracted into data. The function of RPA, which is explicitly developed to bridge the technological gaps between different systems and paradigms, as I will show below, makes it a focal point where these processes of abstraction and subsumption can be observed and analysed. Rather than telling a story of linear continuity, therefore, the diagram and its uses in RPA reveal points of disjunction where visuality serves as a critical vehicle of abstraction.
The robot out of the human
In 2001, a couple of British software engineers, Alastair Bathgate and David Moss, founders of the company Blue Prism, stumbled upon a golden opportunity: to automate the work of filling the gaps in enterprise information infrastructures. Their first client, Barclays Bank, was caught in a conundrum of digital systems. It had implemented ERPs (enterprise resource planning systems) for its finance operations and for its call centre operations but, on the ground, that meant its employees had to work with two different systems and manually transfer data between them. The automation of data collection and processing had inadvertently led to more manual data labour. Bathgate and Moss, with their new company Blue Prism offered a solution – to automate the work of transferring data across systems by using a custom combination of screen scraping, macros and Visual Basic, a programming language that allows users to code by constructing simple diagrams (High, 2019). These diagrams visualise the steps in simple, white-collar tasks in a way that mimics the established visual language of business process notation and is easy to understand by workers with no IT training. Each step in the process is represented by a box in the diagram and each of the steps is linked to the next in a sequence of visualised actions. Users can choose from a number of predefined actions, such as opening a web browser or another application or filling in fields on a webpage or a form, which are then customised to be used in the automation of a specific sequence of tasks (Figure 1).
Example of an RPA tool interface showing the visual programming of an automated process. Accessed from https://digitalexchange.blueprism.com/dx/entry/3439/solution/blue-prism-free-on-prem-trial
These tools, called RPA, sit on top of legacy systems, mimicking human user behaviour and, importantly, making use of key interfacing technologies in digital systems – graphic user interfaces (GUIs) and application programming interfaces (APIs). RPAs interact with these interfaces and transfer data between different systems, making it possible to retain legacy systems while enabling the transfer of data between them. They offer low-code and no-code interfaces that allow users to construct and run automated processes by using the visual language of process diagrams. Bathgate and Moss advertise them as a ‘digital workforce’, promising their clients a fleet of bots that can take over simple, rule-based tasks and implement them with more accuracy and greater speed than human employees.
These tools continue to be widely used, aided by the consequences of the financial crisis that pushed companies to search for cost-saving solutions and banks to look for ways to increase regulation compliance. The COVID-19 pandemic also gave a significant boost to the industry that had already been on the rise for several years, with market research companies projecting a market share of 6.5 billion USD for 2025 (Biscotti et al., 2019) making it one of the fastest growing digital trends.
In contrast to large-scale ‘heavyweight’ automation systems like enterprise resource planning (ERP), RPA does not require the introduction of new IT architecture. RPA is marketed as a solution that ‘democratises automation’ (Blue Prism, 2020; Edwards, 2019; Redbond, 2018) and offers a cheap and easy way to automate specific tasks without requiring a full-scale automation architecture, such as an ERP. This makes RPA a specific type of automation technology that, from its very inception, has developed in a mediating relationship to both human labour and digital infrastructures. RPA does not constitute ‘complete, integrated platforms, either on-premises or in the cloud, managing all aspects of a production-based or distribution business’ (Oracle, 2022).
The rhetoric of ‘democratising automation’ and introducing a fleet of ‘bot workers’ reveals the complex and contradictory relationship between labour and technology at the core of RPA. This relationship serves to reorganise the categories of skilled and unskilled labour. RPA companies describe the tools through the rhetorical figure of the ‘virtual workforce’, which is heralded as a technology that ‘takes the robot out of the human’. Drawing on the original etymology of the word ‘robot’ in the Czech language, where it was crafted by Karel Čapek on the basis of the Slavic stem for both ‘work’ and ‘slave’, this branding problematises the relationship between labour, skill and subjectivity, reinforcing the idea of the separation between conception and execution critiqued by Harry Braverman in the 1970s.
Braverman (1998:78) argued that an important development in the history of scientific management and the optimisation of labour in capitalist production is the ‘dissociation between labour process and skills’, whereby the work process is defined not through the knowledge and experience of the worker but through an abstracted sequence of predefined tasks set by the manager. This abstraction of labour in capitalism as a sequence of actions that can be organised and optimised reframes labour as a malleable medium, rather than an antagonistic subject. The legacy of this separation is evident in the notion of ‘routine tasks’ operationalised in the marketing of RPA, which reinforces the perception that, at least in part, the labour process can be abstracted and standardised to the extent of almost complete separation from individuality, choice and deviation.
Braverman’s work set the beginning of ongoing debates about the hierarchies of skill and control that develop around the introduction of technologies of automation and standardisation in the workplace, which play an important role in the perception of RPA. While his analysis is largely read to imply the deskilling of labour through processes of industrial standardisation and technological innovation, this argument was challenged in the work of Daniel Bell (1973), who claimed that technological innovation generates the need for new skills in the workplace. The narrative of upskilling is one of the key promotional slogans of companies offering RPA. The idea of ‘taking the robot out of the human’, ‘freeing human employees to do more important work’ (Mitchell, 2021) or ‘higher-value work’ (IBM, 2019) is a recurring theme in marketing materials. These more important or higher-value tasks are variously described as management work or customer-oriented work, i.e. highly emotional labour (Automation Anywhere, 2022). Suggesting the perceived limits to automation, these types of jobs reinforce a rigid model of gendered hierarchies of labour within organisations, defined through an opposition between the labour of control and design, on the one hand, and emotional labour, on the other hand (Acker, 1990) where everything in between is seen as fertile ground for overhaul through automation.
The executable diagram
In the early days of scientific management, two enthusiastic followers of Taylor’s (1911) principles, Frank and Lillian Gilbreth, adopted a strikingly visual methodology for labour intensification. Instead of utilising the kinds of time studies that Frederick Taylor adopted to measure how long it takes a worker to perform a task, the Gilbreths used motion and micromotion studies. Their method of micromotion studies made use of the new media of cinema and film to capture the movement of workers on camera and analyse their efficiency. Sat in front of a blackboard with white chequered lines forming a grid behind their back, with a stopwatch in the background, workers would perform their tasks and be filmed. This new technology of motion capture was advertised by the Gilbreth family as a novel and more scientific way to analyse and improve the efficiency of labour and they proclaimed it to be superior to the time studies performed by Taylor (Price, 1989). In some of their later studies, the Gilbreths attached diodes to the fingertips and hands of workers so that they could also trace the trajectories of their movements (Figure 2).
Micromotion study of a female typist. Screen capture from the original movies of Frank and Lillian Gilbreth (2007).
This new use of cinematic technology and the increased interest in capturing the movements of the working body combined earlier experimentations in photographic capture of movements where science and aesthetics intertwined (such as in the work of Eadweard Muybridge) with the new desire to record, analyse and control movement that emerged in the works of scientific management. Films recorded by Frank and Lilian Gilbreth were used not just as a tool to capture the movements of working bodies, but also as a teaching aid to help workers improve the efficiency of their movements by observing, mimicking and correcting their gestures (Mees, 2013; Price, 1989).
These films served as a visual algorithm that provided a diagram and a model for performing work tasks. They not only had the documentary purpose of recording, but also showed the role of technological images in organising the materiality of movement and production. This materiality received a notably literal interpretation in Frank Gilbreth’s work, which later included making 3D models that replicate the trajectories of movement of the workers’ bodies which are made visible by the attached diodes. These visualised 3D-modelled trajectories were set against a 3D grid that helped to quantify them. It is not an overstatement to say that Gilbreth was probably the first person to create 3D visualisation models from the data of his studies. The visualisation techniques he used were essential in analysing the mechanics of work and optimising them (Figure 3).
3D models of work motion trajectories. Screen capture from the original movies of Frank and Lillian Gilbreth (2007).
The Gilbreths did not only film workers as a way to train their movements and organise the workflow. Another important innovation that Frank Gilbreth introduced was the process chart – a detailed visual diagram of how a work task has to be performed and the actions it involves. They developed a special notation system for this purpose called ‘therbligs’ (an anagram of their family name). These process charts are the predecessors of contemporary workflow diagrams widely used in business planning and robotic process automation. In the diagram below, a process chart for loading rifle grenades from Frank Gilbreth’s works prescribes the steps in the process of assembling, checking, and packing rifle grenades in a factory, with different types of operations marked with different symbols. Here, the visualisation of the labour process is more abstract than in the film materials and the body of the worker does not feature that prominently; instead, the focus is placed purely on the processuality and sequence of the workflow (Figure 4).
In the work of Gilbreth, the diagrams served the important purpose of visualising the desired, ‘right way to do a job’ but they also obscured and made invisible other tasks, labour and workers. As Gilbreths’ contemporaries noted, while some labour processes were highlighted as exemplary for the success of the optimisation technique, the Gilbreth method led to the emergence of new, inefficient and suboptimal processes in which all tasks that did not yield themselves to being optimised were outsourced, obscured and invisibilised (Price, 1989). Such regimes of selective visibility are at the core of how the relationship between imagery, technology and control is operationalised and enacted in processes of automation. Images of streamlined processes are not merely functional tools to optimise tasks but also key to reinforcing a narrative of science and technology diminishing the role of human labour or being able to completely dispense with it. From the workers left out of the Gilbreths’ diagrams and films to the role of ‘artificial artificial intelligence’ Turkers toiling in the shadows to power proclaimed hi-tech tools (Irani, 2015) to recent reports of the low-paid human labour supporting the operations of ChatGPT, the interdependence between visualisation and invisibility lies at the heart of the hidden abode that supports the ‘charade’ of automation (Taylor, 2018). This hidden abode of automation is predicated on the antinomies of visuality, labour and technology.
The process diagram plays a key role in reinforcing a regime of selective visibility along the lines of a separation between knowledge and execution in the labour process (Braverman, 1998). The diagram represents and emphasises the abstracted cognitive aspect of the labour process – design, conception and control – which, as Braverman argues, is intentionally alienated from the worker and reinterpreted as a function of management. From the early years of scientific management and the work of Frank and Liliam Gilbreth, this function of the diagram positions it as a tool of the managerial class, a function that was later adapted in the use of business process diagrams in the 1990s. The business process diagram gained prominence in the movement of business process reengineering inspired by Michael Hammer’s Reengineering Work. Don’t Automate. Obliterate (1990), which propagates the importance of managerial control over the process of production in order to optimise it. While Hammer’s title and the spirit of his argument suggest that there is an opposition between automation and optimisation, it is important to remember the astute remark of Alfred Sohn-Rethel (1978:156) that the process of real subsumption of labour under Taylorism ‘physically and technologically … represents the basis and starting-point for the process leading up to the automation of human labour in the precise technical term of the word’.
In the early 2000s, the establishment of the process diagram as a standardised language in itself, codified by the Business Process Model and Notation (BPMN) organisation, the Object Management Group (OMG), and the Organization for the Advancement of Structured Information Standards (OASIS), built on the legacy of scientific management with its visual representation of tasks in time and motion studies. This standardised notation further reinforced the link between visual abstraction and executive function. Here, the executability refers to two intertwined developments. On the one hand, the ideological separation between the design and control functions of management established through scientific management, its ‘executive function’, is reinforced. On the other hand, the incorporation of these diagrams in digital information systems creates another, material dimension of their executability. BPMN diagrams themselves become ‘executable’; the abstract and idealised representations of business processes become rules according to which information is passed between workers and departments in a predefined sequence.
The diagrammatic representation of the labour process in RPA points to continuities in the abstraction of labour and the selective visualisation of skills. The narrative that certain jobs are exempt from automation, perpetuated through the notion of ‘freed’ creative workers in RPA marketing materials, rests on a dichotomy in how labour is conceptualised. On the one hand, through the figure of the ‘digital workforce’, work is presented as a pure abstracted labour process, while, on the other hand, in the figure of freed creative, management and emotional labour, there is an overemphasis on the subjectivity of the worker and its role in generating value. Aneesh Aneesh (2001) notes a similar dynamic in how the notion of skill is influenced by the development of digital technology. He uses the concept of skill saturation to discuss how work rationalisation and innovation in digital technologies create the conditions for certain activities to be more rule-based and, therefore prone to skill saturation, while others are seen as more dependent on subjective judgement. Peter Sawchuk (2006) notes that the contradiction in how capitalist development and technological innovation affect the socially constructed notion of skill and its effects on worker power and autonomy needs to be read through the inherent dichotomy between use value and exchange value in a capitalist system. While the use value of labour tends to create the conditions for cooperation, creativity and the expression of subjectivity, its subsumption under the imperative of exchange value drives an opposite dynamic of standardisation and routinisation of work.
It should be noted that these distinctions are not stable; different types of jobs can be defined as routinised or not through a process that Lilly Irani (2015) calls the ‘boundary work’ of distinguishing innovative labour from menial or routine labour. This unstable boundary is increasingly exploited in the development of RPA as a technology of layered automation, whereby RPA tools are reinvented as platforms that offer plug-and-play machine learning and AI capabilities, as well as through the very logic of low-code programming, which routinises and automates the labour of software development. These platforms underscore the instability of the distinction between skilled and unskilled labour by shifting the perception of which tasks can be automated and which cannot. They offer plug-in AI-powered services like chatbots and sentiment analysis, which contradict the notion of the ‘digital workforce’ as only performing routine tasks and encroach on the contested area of ‘affective computing’ where workers and bots can be alternatively employed to perform actions that are perceived to require more subjective and adaptive responses (Huws, Spencer & Joyce, 2016).
Visuality and autonomy in software labour
In the history of software development, these antinomies of visuality, labour and technology acquire a more complex and contradictory meaning. The low-code programming tools used in RPA allow for a sort of sandbox environment for testing software, combining experiment and control: the visualisation allows for real-time evaluation of the product as it is designed, while offering a user-friendly interface for application development without the hurdles of coding. This poses obvious advantages in an industrial context, shortening significant lags in assembly in the production chain of designing software and graphical user interfaces (GUI) and lowering the need for high-skilled labour. Low-code tools, however, also point to the contradictory development of software programming as not only the paradigmatic figure of innovative labour in the digital economy but also as one that is continuously subjected to self-automation. As Wendy Chun (2005) points out, the history of coding is one of a series of contested loci of control and notions of skilled labour. Chun sees the emergence of coding as indicative of transformations in the medium of computing (shifting from a focus on hardware operation to software programming) and in the hierarchies of labour. The shift from a hardware-centred practice to a software-centred one is accompanied by a gender shift in the constitution of the computing labour force: from female to pronouncedly male. While this development leads to the increased professionalisation of software development it is also embedded in a tendency for its increased abstraction and the increasing importance of the role that symbolic languages play in this abstraction.
The low-code tools used in RPA are part of the ambivalent trajectory of the progressive abstraction of software programming. The role of visualisation in programming and automation has been obscured by a focus on code as a cultural and political technology of translation and ‘black-boxing’. However, diagrams and flowcharts constitute an important part of the development of computing, from the diagrams used to visualise algorithms in the work of Goldstine and von Neumann (1947) to later transformations of the flowchart in software engineering (Morris & Gotel, 2012, 2006). Diagrams and flowcharts occupy a central place as instruments of scientific knowledge where they are used to visualise invisible flows, such as electric currents, and to represent sequences of actions in mathematical calculations and algorithms.
This convergence of software automation and business process management in RPA is not an accidental occurrence in the history of software labour but, rather, signifies an inherent contradiction that links the development of software programming to its increased abstraction and automation. In the early 1980s, James Martin (1982) wrote that the ultimate goal of application developers should be to automate their own work, arguing that programming languages that are more accessible for end users will not only cut the cost of labour but will also help bridge a gap between the project cycle of business management and that of software development.
The desire for more efficient coding also underpins the movement for agile programming and rapid application development. ‘The Manifesto for Agile Software Development’ (Beck et al., 2001) articulated calls for coders’ autonomy through the idea of a quick turnaround cycle of software production with an emphasis on prototyping and sandbox environments (Bulajewski, 2013; Lennon, 2018). The manifesto adopts the principles of the agile methodology introduced as part of just-in-time production in neoliberal economies – an emphasis on the rapid feedback loops that force the production process to constantly adapt to the demands of the market (Moore, 2018). Rapid application development (RAD), in particular, evolved as a type of agile programming, which departed from the ‘waterfall’ method of project management, widely used at that time, and focused instead on speed of delivery, with user feedback based on testing the prototypes (Boehm, 1988; Martin, 1982). The agile programming movement reclaimed key principles of scientific management – the visualisation of production processes and the push for intensification and optimisation of the labour process, in a paradoxical way. In contrast to the real subsumption of labour under the control of management, the use of visual tools and rapid application development was articulated as a path to autonomy for the coders, distinct from a timeline of the production process shaped by the logic of business management. The ability to further automate their own labour and accelerate the production timeline was seen by the authors of the manifesto and its signatories as a way to claim autonomy and control in the RAD community. This seemingly paradoxical relationship points to an inherent contradiction at the heart of cognitive labour and its embeddedness in capitalist production. While the increasing reliance on cognitive and creative skills for the generation of value empowers workers and constrains the incentive for capital to deploy Taylorist methods for real subsumption of labour, the products of this creativity confront the worker as ‘dead knowledge’ (Vercellone, 2007) – i.e. the intellectual and creative power appropriated by capital which is used to optimise profit at the expense of real workers’ control.
However, the use of low-code programming and visual tools in RPA, where the push for automation is loaded with implications that coding can be routinised, underscores the instability of the distinction between routinised and creative or skilled labour. It gestures to the specific context of automation and control in the progressive digitalisation of industries. This context is informed on one hand by the medium of digital systems and the affordances for further automation that they enable. On the other hand, however, the contradictory position of software development speaks to the interdependence between abstraction as epistemology and abstraction as a mode of social organisation, as suggested by Alfred Sohn-Rethel (1978). He argues that we should understand the emergence of abstract thinking in scientific thought as related to the social abstraction of exchange value and that this historically conditioned affinity contributes to the tendency for technology and scientific thought to become subsumed under the imperatives of capitalist profit and to reinforce the division between menial and intellectual labour.
Platformisation of the diagram
Alongside its role in the conflicted dynamics of labour automation and control, low-code RPA is increasingly integrated into cloud platforms. There, the workflow diagram is incorporated within economies of data extraction, software standardisation and ‘hyper-automation’. Companies are offering RPA as part of layered automation that combines low-code programming tools, machine learning and AI. These integrated platforms include access to predefined automated processes that can be copied and modified by the users: for instance, a process for filling in an Excel table and performing calculations in it. Big tech companies like Amazon Web Service and Microsoft Azure are offering low-code RPA and third-party machine learning and AI tools that help accelerate the automation of processes – including IBM’s Watson, Google’s TensorFlow and ABBYY’s optical character recognition (OCR), as well as process mining. But traditional RPA companies like Appian, Blue Prism and Automation Anywhere are also rebranding themselves as platforms.
Within the ecology of these platforms, the process diagram participates in important dynamics of reorganisation of control and the transition between labour abstraction and the extraction of data as a new source of value generation. The incorporation of low-code RPA in hyper-automation platforms points to the emergent tendency of consolidation of the means of automation within a section of big players who leverage their power over the market through cloud-based Platform-as-a-Service (PaaS) infrastructure that is becoming a key technology for the accumulation of data and capital. Companies, such as Oracle (Mueller, 2018), SAP (offering the low-code platform Mendix also offered by IBM), Salesforce (Eammano, 2019) and Appian (Calkins, 2019) are currently leaders in PaaS low-code products, incorporating them into their already established digital services for industrial automation and management.
Louise Amoore (2018) and Anna Munster and Adrian Mackenzie (2019) argue that the role of visualisation as a technology of control (which can be read in relation to low-code diagrams) is complicated by its incorporation in the platform and the cloud as infrastructures for accumulating data, power, and capital. While Amoore discusses the emergence of the cloud as an experimental technology, which enables selective visualisation and obfuscation, Munster and Mackenzie suggest that the convergence of visual data and massive platforms for its storage and analysis leads to new heterogeneous logics of control and extraction by multiplying the human and non-human agents who ‘see’ and use digital data. The proliferation of devices, sensors, algorithms and tools for image processing, however, solidifies the perceptual asymmetries of ‘platform seeing’ where the transversal, diagrammatic affordances of visuality in the platform make it impossible to observe the process of image production, extraction and manipulation in their entirety. Platforms complicate the locus of control by creating an environment where ‘modularity and power are negotiated between a core unit with low variability and heterogeneous components of high variability’ (Plantin et al., 2018:298), a tension that becomes a vehicle for further automation and extraction. Within the architecture of the platform, the diagram occupies a critical role in enabling this tension by combining the standardisation of programming with the apparent ease of customising and adapting RPA templates. This tension is productive of new forms of concentration of ownership and monopolisation (Narayan, 2022; Srnicek, 2017) while also enabling the transitions and frictions between the subsumption and abstraction of labour and practices of data extractivism (Couldry & Mejias, 2019).
Although we cannot claim a linear transition between practices of labour subsumption and the economies of data extraction, process mining tools are often articulated as technologies that smooth over the differences between different agents (and subjects) of labour, treating human workers, bots and other digital systems alike as data generating entities (UiPath, 2020a). For instance, UiPath promises to further optimise all processes through its ‘intuitive’ process mining by visualising ‘how your robots, people and systems work together’ (UiPath, 2020b). Data traces are analysed through third-party services for process mining and workflow mining where the introduction of an RPA tool acts as a means for the further acceleration of automation. The data extracted from different applications are converted into event logs that structure temporally a sequence of actions and can then be visualised as a process and optimised to avoid bottlenecks and inefficient actions. UiPath’s website puts particular emphasis on the visualisation of the ‘ideal’ process and its current state as a way to ‘reveal’ the hidden insights through data analytics (UiPath, 2020a) underscoring the complex role of diagrams and visualisation as drivers of continuous automation, subsumption and extraction. Optimised processes are then fed back into the operations of the company, either through revised RPA or by requiring workers to perform their tasks differently. The incorporation of process mining and AI tools in platforms for hyper-automation and intelligent automation shows the extent to which different notions of control and organisation of production can co-exist and be imagined as a continuum rather than an opposition. While RPA itself clearly reproduces Taylorist ideals of managerial control in the subsumption of labour, machine learning and AI tools operationalise the current context of digital media and big data in shaping the possibilities of decision-making and imposing new machinic logics of control and valorisation (Pasquinelli, 2015).
The focus on data as the main object of intervention underscores a key problem in the analysis of digitalisation and the question of control and power in automated systems: the extent, to which technological innovation introduces radically different modes of control. While the marketing image of RPA presents it as a simple rule-based automation, bots are part of the technologies that enable new forms of ‘digital Taylorism’ and scientific management (Moore, Upchurch & Whittaker, 2018) whereby software robots participate in the complex relationship between recording, control and further automation. Bots are not only seen as a possible replacement for human workers, but they are also a crucial technology for the digitalisation of business operations in ways that underscore the increasing entanglement of robotic processes and the operationality of digital media. More than being regarded as stand-alone products and technologies, bots are increasingly valued because of the way they facilitate the generation of structured digital data in the form of logs and filled forms, their ability to feed data into AI and machine learning applications, and their capability for enabling the transfer of data and interoperability between different systems (Suri, Elia & van Hillegersberg, 2017). They scan, track, quantify, and monitor the movement and performance of workers within contemporary regimes of ‘surveillance capitalism’ (Zuboff, 2019) while, at the same time, continuously generating data records of their own actions that serve as the resource for new machine learning, automation, and AI applications. RPA in this sense embodies contradictory notions of control and power in automation: the tension between continuity in the development of scientific management and the new affordances of control enabled through the medium of digital data.
Conclusion
The growing interest in low-code programming in recent years extends to various software products and tools that have adopted diagrams and charts as an easy modular way of programming, like the game development console Unreal Engine, designer software Touchdesigner, the manufacturing automation platform Kissflow and many other tools spanning the whole spectrum of industries – from creative to factory production. This resurgence in the instrumentalisation of the relationship between visuality, digitality and control reveals the conflicted position of the ‘technical image’ in digital capitalism. Its position is predicated on the combination of several important influences in the development of process charts and diagrams as tools for the organisation and automation of the labour process.
The first influence is the continuing relevance of the aesthetics of managerial abstraction, which not only draws a direct line of continuity from the early years of scientific management and the use of diagrams for analysing and optimising work tasks but also shapes the instrumentalisation of (in)visuality in capitalist production. This instrumentalisation is enacted through the separation between conception and implementation, noted in Braverman’s (1998) critique of scientific management and the hierarchisation of labour into low-skilled and high-skilled categories. In both of these processes, technology is instrumental in enacting the visibilities and invisibilities of labour through the streamlined abstraction of the diagram which presents an idealised version of the labour process and obscures the frictions, inefficiencies and incongruities behind it. Against the frictionless image of RPA ‘taking the robot out of the human’ that companies are promoting, one of the persistent struggles remains to secure the consent and cooperation of workers who are worried about losing their jobs or losing control over their work (Seiffer, Gnewuch & Maedche, 2021). This underlying antagonism speaks to the continuing relevance of labour process theory, not just in understanding the mechanisms of automation but also in seeing it as a frontier of the persistent struggle of labour against its abstraction and subsumption. The aesthetics of selective visibilities afforded by the diagram serve as a vehicle for the real subsumption of labour while, at the same time, aiding the ideology of technological determinism to hide the inherent limitations of automation, a point succinctly made by Lily Irani (2015:231).
Second, these low-code visual tools for automation and programming function as devices for the accumulation of structured data, through process logs generated with each use. These structured data produce a second, hidden level of automation by enabling other software systems to analyse and further automate processes. The invisibilities in this layered automation add another dimension to simple, easy-to-use tools like RPA. The appeal of code-less programming and the allure of the idea that everyone can create their own fleet of bots that perform the tedious tasks of clerical digital labour hide the intersections of data extraction and black-boxing in current low-code tools. Not only are the low-code tools based on predetermined functions that users can modify to a minimal extent, but they are also increasingly embedded within the economies of digital cloud platforms. These cloud platforms perpetuate the concentration of resources and digital means of production in a handful of technological companies. Much like the ‘operational’ or ‘operative’ images, the process diagram resides in the ambiguous tensions between visibilities and invisibilities, where the aesthetic of efficiency and streamlined processes enables the invisual accumulation of data and acceleration of automation. The process diagram does, however, underscore the key position of labour in the production, use and resistance against the operationalisations of visualisations.
Finally, the example of the agile programming movement suggests a potential for subverting the oppressive and exploitative uses of technologies of visualisation. While the movement borrows from the language of business management, its underlying ideas of workers’ autonomy, experimentation and easier access to the tools for programming can perhaps serve as a model of emancipatory reappropriation of technology.
In the course of the focus groups at Bank X, it transpired that while the company had introduced a culture of creativity and experimentation, this remained confined to just one of the departments focused on UX design. At the same time, the accumulation of knowledge, experience and ideas among the workers in other departments who had, in the course of their employment, developed ingenious strategies for organising and sharing knowledge among themselves and being efficient (in the sense of saving their own time and effort) was gradually being erased. In the course of the conversations, one of the powerful images that emerged was that of the red notebook of a long-term employee, a notebook where he had collected and organised key information about organisational procedures, rules and tips for addressing different inquiries. The red notebook stood in contrast to the new AI-powered knowledge base systems introduced by management that could be queried in combination with the RPA bots. The sense of pride and ingenuity that this red notebook embodied served as a potent symbol to guide our own analysis and recommendations and to inspire a vision of a workplace where existing forms of knowledge and experience are incorporated within the design of new IT systems and where experimentation and creativity are extended to all workers, not just as abstract principles, but by freeing time, reducing workloads and allowing workers to move across teams and be actively in control of their own career development. This example shows the potential of visuality, the ideological power of aesthetics but also the visual as a ‘map’ (Beverungen, Beyes & Conrad, 2019), to be employed in ways that do not reduce the power of workers but, instead, inform new horizons of achievable utopias of automation.