Digital Taylorism: How AI Reshapes Control in Service Work
This article is adapted from earlier academic work and has been edited for a general audience.
There is a version of the AI-and-work story that goes like this: AI automates routine tasks, freeing human workers to focus on higher-value, more creative, more meaningful work. It is a tidy narrative. It appears in corporate strategy decks, consulting reports, and technology press coverage with striking regularity. And it is not entirely wrong. But it leaves out something important about what happens to the work that remains.
When AI enters a workplace, it does not simply remove tasks and leave everything else intact. It changes how the remaining work is organised, monitored, evaluated, and controlled. The question is not only which tasks AI can perform. It is what kind of working life AI creates for the people still doing the rest.
When the manager is the machine
One of the less visible consequences of AI adoption is the transfer of managerial functions from people to systems. Scheduling, task allocation, performance monitoring, quality evaluation: these are decisions that were once made by human managers, often imperfectly, often with significant discretion. Increasingly, they are made by algorithms.
In customer support, this might mean AI determining which agent receives which query, how long they should take, what scripts they should follow, and how their performance is scored. In logistics, it might mean automated scheduling that adjusts shift patterns in real time based on demand forecasts. In content moderation, it might mean algorithmic systems deciding what volume of material each worker reviews and flagging deviations from expected throughput.
The sociologist David Spencer characterises this broader shift as a form of digital Taylorism. The reference is to Frederick Taylor's early twentieth-century scientific management, which sought to break work into its smallest components, study each one for efficiency, and return optimised instructions to workers. Spencer's argument is that AI performs a contemporary version of the same process: managerial and experiential knowledge is extracted, embedded in technology, and returned to workers as predefined instructions. The worker's role shifts from exercising judgement to following the system's logic.
This is not inherently harmful. There are contexts in which algorithmic coordination improves consistency, reduces bias in task allocation, and makes workloads more predictable. But it changes the nature of the relationship between worker and work. And the consequences of that change are uneven.
The skills that stay, the skills that go
A common claim about AI in the workplace is that it will drive widespread upskilling: workers freed from routine tasks will develop more sophisticated capabilities and move into higher-value roles. The evidence is more complicated than that.
AI does not simply displace workers or leave them untouched. It alters the task composition of jobs, automating some activities while creating others. The effects are real but unevenly distributed. Productivity gains from AI tools tend to be largest among less experienced workers. In customer service, AI helps newer agents handle problems they would previously have escalated, compressing the performance gap between junior and senior staff.
On the surface, this looks like democratisation. But there is another way to read it. If AI narrows the gap between experienced and inexperienced workers by providing everyone with system-generated guidance, then the experiential knowledge that once distinguished senior workers becomes less visible and less valued. The skills that remain in demand shift toward what might be called algorithmic literacy: the ability to interpret system outputs, detect errors or bias, and determine when human judgement should override automated suggestions.
Meanwhile, the conditions for developing deep experiential expertise may be eroding. When interactions are scripted, when decision pathways are predetermined by algorithmic workflows, and when performance is measured against system-generated benchmarks, there are fewer opportunities for workers to build the kind of knowledge that comes from navigating ambiguity and making independent judgements over time. Spencer's digital Taylorism framework captures this tension precisely: the system benefits from the knowledge that experienced workers have accumulated, but the structure of the work may prevent the next generation from accumulating it.
AI is also generating genuinely new categories of work. Testing, training, and supervising AI systems are roles that did not exist a decade ago. Customer service workers are already contributing training data through their interactions, feeding the very systems that may reshape their roles. AI-related skills carry significant wage premiums. New work is emerging. But the pathway from routine service work to AI-adjacent roles is not automatic, and organisational support for making that transition remains patchy. Many workers rate the quality of their AI training poorly, and the kind of sustained investment required to reskill a workforce is difficult for smaller organisations to replicate.
The result is a simultaneous pressure for upskilling and deskilling within the same organisations. That is a harder story to tell than "AI frees workers for better tasks," which may be why it is told less often.
Visible workers, invisible logic
There is a further dimension to this shift that deserves attention: the asymmetry of visibility it creates.
Ifeoma Ajunwa describes this as the "black box" at work. In algorithmically managed environments, workers are highly visible. Their activity is tracked, their performance is scored, their productivity is measured against benchmarks generated by the system. But the logic by which the system makes its assessments is often opaque. Workers can see the outcomes of algorithmic decisions, a performance score, a task assignment, a scheduling change, but not the reasoning behind them.
This opacity matters because algorithmic management systems are often justified as reducing human bias. Decisions made by data, the argument goes, are more objective than decisions made by individual managers with their own preferences and blind spots. But fairness in AI is not a purely technical property. It is shaped by organisational goals, power relations, and the assumptions embedded in system design. Algorithmic management can reproduce or amplify discrimination when biases present in training data or design choices are reflected in automated decisions. Systems that appear neutral can entrench historical patterns of disadvantage by embedding them in processes that no individual decision-maker can easily see or challenge.
In customer support, these dynamics are particularly concentrated. AI systems may determine which agents receive the most complex or most rewarding queries, how performance is benchmarked, what constitutes an acceptable resolution time, and how deviations from expected patterns are flagged. Workers operate within these frameworks, often without meaningful visibility into how the parameters are set or how they might contest an adverse outcome.
The design choice that gets overlooked
It would be easy to read all of this as an argument against AI in the workplace. That is not the point.
Research by Wang and colleagues offers an important counterpoint. Their work examines AI-enabled conversational agents functioning as "digital employees" in frontline service. They find that when these systems are designed to redistribute work rather than replace it, they can enhance workers' sense of job control and professional identity. The technology is the same. The design intent is different. And the outcomes for workers diverge significantly.
This distinction is at the heart of the issue. AI systems that automate task allocation, monitor performance, and score productivity are not inherently exploitative. But when they are implemented without transparency, without mechanisms for challenge, and without attention to how they reshape the experience of work, they reproduce a very old pattern in a new technological form: the concentration of knowledge and control at the top of an organisation, and the reduction of discretion at the bottom.
Spencer's digital Taylorism is not a prediction about the future. It is a description of a tendency already present in how many organisations deploy AI. Whether that tendency defines the next decade of work, or whether it is tempered by more thoughtful approaches to design, governance, and worker involvement, depends on choices that are being made right now.
The technology does not decide. The organisations using it do.
References
Acemoglu, D., Autor, D., Hazell, J. and Restrepo, P. (2020) AI and jobs: Evidence from online vacancies. NBER Working Paper No. 28257. https://doi.org/10.3386/w28257
Adams-Prassl, J., Binns, R. and Kelly-Lyth, A. (2022) 'Directly discriminatory algorithms', Modern Law Review, 86(1), pp. 144–175. https://doi.org/10.1111/1468-2230.12759
Ajunwa, I. (2020) 'The "black box" at work', Big Data and Society, 7(2). https://doi.org/10.1177/2053951720938093
Brynjolfsson, E., Li, D. and Raymond, L. (2025) 'Generative AI at work', Quarterly Journal of Economics. https://doi.org/10.1093/qje/qjae044
Eubanks, V. (2018) Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor. New York, NY: St. Martin's Press.
Jarrahi, M.H., Newlands, G., Lee, M.K., Wolf, C.T., Kinder, E. and Sutherland, W. (2021) 'Algorithmic management in a work context', Big Data and Society, 8(2). https://doi.org/10.1177/20539517211020332
Selbst, A.D., Boyd, D., Friedler, S.A., Venkatasubramanian, S. and Vertesi, J. (2019) 'Fairness and abstraction in sociotechnical systems', in Proceedings of the Conference on Fairness, Accountability, and Transparency. New York: ACM, pp. 59–68. https://doi.org/10.1145/3287560.3287598
Spencer, D.A. (2017) 'Work in and beyond the Second Machine Age', Work, Employment and Society, 31(1), pp. 142–152. https://doi.org/10.1177/0950017016645716
Tambe, P.B. (2026) 'Reskilling the workforce for AI: Domain expertise and algorithmic literacy', Management Science, 72(1), pp. 515–537. https://doi.org/10.1287/mnsc.2022.03968
Wang, Y. et al. (2025) 'AI-enabled conversational agents as digital employees in frontline service', Information and Management. https://doi.org/10.1016/j.im.2025.104099