Picture a modern, high-tech warehouse. Here, human managers occupy a sort of fallback position: They still exist and still play a kind of supervisory role, but workers receive direction and have their performance monitored through the devices with which they interact. This is algorithmic management, and it can be extremely efficient because a complex software algorithm can direct workers more minutely than even the most micromanaging supervisor.
A packer in such a warehouse will find themselves at a workstation with a screen and a scanning gun: They scan a package, the screen tells them exactly how to pack it, and a machine spits out pieces of tape of the right length. Precisely how long this operation takes is fed into a monitoring and evaluation algorithm, which will calculate whether they are meeting the rate: the desired, often extremely intense, work rate. If they fall below the rate, they will get a warning and sometimes a visit from a manager following the output of their own algorithm: Go and exhort that packer to pack faster. At some point, determined within the black box of the algorithm, their manager receives a recommendation to sack them.
That packer’s boxes are sent to them by a picker, who pushes a trolley around a warehouse on the orders of another algorithm. This algorithm is efficient because it allows boxes to be stowed randomly when they enter the warehouse, which takes much less time—it’s called chaotic storage—and then pickers are given maximally efficient routes to follow to pick up their items. They’re not told those routes: They are given the location of the next item and a timeframe in which to reach it. As one sociologist reports, it “is a strange sense of time that the device produces, in which the worker’s ability to control their own time is repeatedly reduced to the next twelve seconds in which they must find and pick their next item.” In the near future, they may get instructions on precisely how to move their hands to efficiently select their item, transmitted through the vibrations of special wristbands that Amazon patented a few years ago.
If the order gets mispackaged anyway, and a customer calls to complain, they might reach one of Amazon’s call centers that uses Cogito, an algorithmic management tool that monitors calls and gives workers real-time instructions on their speaking pace and tone of voice. If you sound tired, it apparently displays a little coffee-cup icon. Their results are quantified and fed into their manager’s evaluation algorithm.
In my first post in this series, I described a perennial concern of political theorists and political economists: that whatever else we make at work, we also make ourselves. The way that we work shapes us, allowing us to develop some skills and habits, and stripping us of opportunities to develop others. So, among the many questions we ought to raise during this latest transformation of the workplace, we should ask whether the rise of algorithmic management seems likely to enable or frustrate the development of capacities we have reason to value—in particular, the loose set of what we might call democratic capacities, those skills and habits required for ruling ourselves as equals.
In trying to answer this question, I want to gesture at a few of the most significant changes to the experience of work under algorithmic management, the first and most striking of which is the depersonalization of authority involved in these workplaces. This is perhaps what feels newest about management by algorithm—it can seem like an extraordinary shift in the very structure of work has taken place and that you’re no longer working for another person at all. Of course, that’s not true. But what is true is that your exposure to authority isn’t personalized in anything like the same way as in a traditional workplace. Maybe the most obviously political feature of a traditional workplace is that you are under the power of another human being: Even if they’re nice, you have good reason to want them to remain nice. There are constraints on what they can do to you (from employment law, from the demands of the market, from their own bosses or shareholders), but your boss can make your life go well or make it go badly—you’re personally dependent on the will of another. This interpersonal relationship of command and obedience inspired a lot of older worries about the hidden curriculum of the workplace. You find nineteenth-century republicans writing that through obedience “to the caprice of he who pays the wages,” deference, servility, adaptation to the whims of those with power over you are all encouraged—imperfectly, but encouraged—while the capacities of self-direction are left to wither on the vine.
Workers in highly algorithmic workplaces still have bosses, of course, but to the extent that their management is performed by algorithm, their relationship to those bosses isn’t strategic. They don’t have to flatter or toady up to their supervisor, and there’s no point worrying what an algorithm thinks about them. They can’t pretend to laugh at an algorithm’s bad jokes or try to keep on its good side in the hope of better treatment. It will spit out instructions and monitor productivity—it will do what it’s been programmed to do—without fear or favor. To this extent, that worry about personal dependence disappears.
But in place of this pressure to routinely defer to another human being isn’t some measure of genuine control over one’s work. There’s no option but to defer to the impersonal authority of the algorithm. In the literature on these workplaces, you read again and again how difficult it is for workers to challenge (or even seek an explanation for) the operation of that algorithm; they are often met with the fiction that the algorithm is objective and unchallengeable at some fundamental level. In some sense, it’s true: Their human supervisors may not understand the complex algorithms that run their day any better than the workers themselves. If you think you’ve been wronged, there’s no human source of that wrong. The passivity, even fatalism, encouraged by this depersonalization of authority can be seen in the words of the manager quoted above: “We didn’t fire you, the machine fired you because you are lower than the rate”; but it can also be seen in the advice Armin Samii, the Uber Eats rider, was given by his colleagues to “trust the system” and “be at peace with it.”
If habitually deferring to another human being provides a poor training ground for democratic self-direction, then habitually obeying an algorithm you are told is unchallengeable doesn’t seem like an obvious improvement. Is it better to incentivize passivity because someone else is making the decisions or because the decisions are being made—it seems—by nobody in particular? For their part, Amazon claims that human supervisors have the ability to override the recommendations of the algorithm when it comes to discipline, a response mirrored in various political proposals to keep a “human in the loop” when it comes to serious decisions like hiring and sacking. For current purposes, this isn’t exactly reassuring: Trust the system and learn to accept that decisions emerge from a black box, but you’d better keep in your remaining supervisors’ good books just in case they might overrule that system when it really matters.
Next is the place of judgment and deliberation in highly algorithmic workplaces. I haven’t laid out a neat list of democratic capacities here, and I’m not sure I could do so if I tried. But some of the capacities relevant to democratic life are epistemic, in at least the sense that to make decisions together, as equals, requires us to reason with each other, to persuade and convince, to make sense of the processes that affect us and the consequences of possible decisions, and to interpret and reinterpret the rules and norms by which we live.
Since at least the beginning of the industrial revolution, the division of labor between those who conceive plans and those who execute them has threatened to rob workers of reason to exercise these capacities. If your job is essentially to follow someone else’s orders, then there’s no incentive to understand the wider process in which your work plays a part, let alone to try to persuade anyone that you could improve that process. But this, of course, isn’t ever quite how it goes. Even fairly routinized jobs in authoritarian workplaces require workers to use their judgment, develop heuristics, and work out what needs to be done to reach their goal. As readers may know from experience, the people actually completing some task at work will often have—and have reason to develop—a level of knowledge about their tasks that the people in charge will lack.
In this sense, even quite authoritarian traditional workplaces allow workers opportunities to develop and exercise some kinds of judgment over work tasks: There’s some level of autonomy there, because instructions can only be so fine-grained and so tightly monitored.
Algorithmic management can be extremely efficient, but only when tasks are routinized to an extraordinary degree, broken up into tiny chunks with easily quantifiable results, and monitored very closely but crudely. This sometimes looks like a warehouse worker being given twelve seconds to move down an aisle and retrieve a box, but administrative work is no less capable of being transformed in this way. A former Amazon human resources worker describes her experience of this routinization: “In the warehouse, [it’s] how many boxes you can scan in an hour, a minute,” she tells an interviewer:
HR … is the same thing. How many cases can you work? Not a single question was: Can you save this person? It’s just: “How fast can you do this case?” At times, the department felt more like a “robot reviewing a case” than an HR team.
Couple this with the opaque nature of the algorithm, and the picture looks fairly bleak. Even if there were time to do so, there’s often simply no way for an employee to discover precisely how their instructions fit into a wider process or a longer-term plan. At least in the form it has taken in sectors like logistics—which is to say, in sectors where it has been enthusiastically adopted over the last decade—algorithmic management risks squeezing these opportunities for judgment and deliberation to an absolute minimum.
Editor’s note: This is an extra edition of the Perspectives on Democracy series for September 2026.
Orlando Lazar
Orlando Lazar is Lecturer in Philosophy at King's College London. His research focuses on work and the workplace, power and domination, and economic democracy. Recent work includes “Taking it Home With You: Work, Free Time, and Domination” (CRISPP, 2025) and “Micro-domination” (European Journal of Political Theory, 2023).
