Is AI in the Workplace Making You Disappear?
- 8 hours ago
- 12 min read
When AI decides not only how we work, but who we must become in order to stay employed.
Many of us know the story of the monk who repeatedly draws water from a well by hand. Someone comes along and points out that he could make the process much more efficient, and suggests ways to optimize the task. The monk responds:
“Why would I do that, for if you make one task quicker to finish, you simply create more time to fill with more work.”
Organizations are constantly motivated to assess “How can we do this better and faster?” This is a legitimate question for both employers and employees.
With widespread use of modern artificial intelligence making its way onto the work floor, the accelerating pace of productivity has given both employees and employers a sense of vertigo. Yet many of us fail to ask, “When does “better and faster” become an insatiable beast?”
Buddhist teachings include the concept of applying the right amount of effort, rather than simply maximizing effort—the Buddha’s “lute strings” analogy says strings that are too tight or too loose don’t produce the desired sound.
Still, most employees find the process of negotiating this balance to be quite a challenge. Would you dare to ask your employer whether it is ok to take it a little easier since you are now able to churn out more work than ever with help from AI? And, if so, how do you think your manager would react? Here are two possible organizational responses:
A) “Great. You saved 90 minutes. Use the time to rejuvenate, learn a new task or method, help someone out, or finish your day earlier.”
B) “You can do that in 30 minutes now? Great. What else can you take on?”
Which response do you think will be more likely in your organization? Which one would you prefer?
When organizations automatically translate increased efficiency due to the use of AI into an opportunity to squeeze more productivity out of their workforce, or even as a downsizing opportunity they neglect to consider the possibility that their employees are increasingly seeing AI as a threat rather than as a work productivity assistant.
A major reason that employees might view AI in a negative way stems from the fact that its rollout within the organization is often implemented without a clear, transparent, and well-conceived plan. Without a clear, well-conceived and transparent strategy, the introduction of AI into the work environment can create a paradox:
Technology that was originally intended to increase productivity may instead generate uncertainty, resistance, and even greater pressure on employees.
The perspectives of an organization’s employees have a significant influence on their employee attitudes both toward their work and toward the organization itself.
Let’s look at some examples of the influences that AI can have on employee perspectives that have predictive value regarding their sense of engagement and personal growth versus their feelings of disengagement, inadequacy, and eventually even professional burn out.
As you read through these alternative perspectives, consider which one reflects your predisposition.
1. Control and autonomy
This connects directly to job control, autonomy, psychological safety, and burnout.
A) “AI gives me back time I can use to learn, create, solve meaningful problems, or work in ways that energize me.”
B) “AI gives my organization back time it can use to increase my workload, raise performance expectations, and fit more work into the same day.”
2. Self-efficacy and competence
This one is extremely influential. Does AI make people feel more or less capable?
A) “I couldn’t have done this myself. AI helped me accomplish something difficult.”
B) “AI did most of this. I don’t even know whether I’m actually good at my job anymore.”
3. Fear of failure
This one is not just about whether AI makes employees feel more or less competent, but whether AI influences who the employees are likely to perceive as responsible for their failures.
A) “AI helps me accomplish difficult things, and when something goes wrong, I can learn and improve.”
B) “AI changes how the work gets done, but I’m still blamed when the outcome is unsatisfactory.”
4. Self-worth contingent on achievement
Research on performance-contingent self-esteem is particularly relevant. Ferris and colleagues¹ found that the relationship between self-esteem and job performance depends on how closely employees associate their self-esteem on workplace performance. Technology is experienced as an extension of the self.
A) “AI makes me better at my job without making my contribution feel less important.”
B) “If AI can produce the work I used to be praised for, what’s left for me to be good at?”
5. Sense of purpose and meaningful work
What happens when AI removes parts of the job that were difficult—but also deprives employees of a sense of meaningfulness? This is particularly powerful because employee may have a lower workload, but experience less desirable psychological outcomes. AI does not simply reduce the effort it takes to accomplish the work—it also has the potential to siphon off the source of the meaning that was inherent in the work.
A) “AI takes away the tedious parts of my job so that I can spend more of my time doing the difficult, meaningful work that makes me feel as though what I do really matters.”
B) “AI is taking over the difficult parts of my job that made my work feel meaningful, leaving me with the easier tasks that seem to matter less.”
6. Social comparison
Because of AI, your personal benchmark isn’t your own past performance; it’s everyone else’s AI-enhanced performance. You could find yourself engaged in a productivity arms race: Once everyone else starts optimizing, not optimizing becomes psychologically unsettling.
A) “The use of AI by others shows me what’s possible; I can choose what is worth adopting.”
B) “The use of AI by others determines what I am now expected to produce.”
Because AI doesn’t merely change the amount of work. It potentially changes:
how much control we have
why we work in a particular field
what makes us feel competent
where we derive our sense of self-worth
whether our work feels meaningful
how we compare ourselves with others
whether we feel safe enough to call it a day
Now sum up your As and Bs and see which way you are leaning.
The higher the ratio of As to Bs, the more you tend to have a sense of engagement and thriving in your job.
The higher the ratio of Bs to As, the more likely you are to feel that AI is influencing your productivity in a way that lowers your level of engagement and might even lead to feelings of professional burnout.
If you belong to the group that tends to favor perspective B over A, you are not alone, and the members of your group are rapidly increasing in number.
Globally, people are struggling with low engagement and mixed feelings of well-being at work. Data from the Gallup State of the Global Workplace Report show that only 20% of workers are engaged. Disengagement costs the global economy an estimated $10 trillion in lost productivity.
Can we blame it on AI? AI isn’t inherently responsible for reduced engagement — but poor management of AI is definitely a significant factor. As Gallup highlights, organizations see an AI-related improvement in engagement only when leaders introduce AI tools in a transparent way, with clear expectations, meaningful work, giving employees a direct voice in how technology can improve their day-to-day lives.
Sounds like a recipe for success, doesn’t it?
Many organizations, however, do not follow this path. Worse yet, they may adopt a different approach to managing employees’ attitudes toward work — one that is not genuinely aimed at increasing engagement, but purports to do just that. And that is where the real danger lies.
Organizations are attempting to counter the engagement crisis by shifting people management away from humans and toward algorithmic control technologies (also known as algorithmic management). Companies deploy data analytics, automated tracking, and machine learning to replace or augment traditional supervisory roles.
This introduces again other perspectives related to AI in the workplace, but before I introduce them to you, I need to explain how the use of AI within organizations is influencing these perspectives. Organizations are increasingly adopting:
Algorithmic control.
Algorithmic control in organizations is the use of computer software, data, algorithms, and automated systems to direct, monitor, and evaluate workers.
Kellogg et al. explained in their 2020 paper² that organizations used to be managed through technical and bureaucratic control. Technical control uses physical technology and machinery to monitor and direct work steps. Bureaucratic control uses rules, policies, and formal authority structures to shape employee behavior and guide daily operations.
Now, instead of human supervisors managing the daily tasks of their workers, algorithms assign work, track performance metrics, and issue rewards or penalties. Algorithmic control, or algorithmic management, operates through six distinct mechanisms²:
Recommend: Suggesting specific actions or routes to employees.
Restrict: Limiting worker choices or access to certain information.
Record: Continuously tracking data like location, keystrokes, or speed.
Rate: Instantly scoring performance via predictive data and customer feedback.
Reward: Automatically assigning bonuses, shifts, or ideal tasks to top performers.
Replace: Automatically shifting tasks away from or locking out underperforming workers.
Kellogg et al. stated that many algorithmic control technologies are sold on the premise that they improve decision-making, coordination of work and organizational learning. However, they are increasingly used to control employees’ behavior.
Using this kind of AI, organizations attempt to not only regulate what employees do, but also what they want their employees to become, and therefore the psychological danger is not simply:
“I am being controlled.”
It is:
“I increasingly feel that my ability to act independently of the system is ineffectual.”
This evolution, from controlling the behavior of employees to controlling the conditions under which employees can act can result in the following:
powerlessness
learned helplessness
anxiety
hypervigilance
self-censorship
loss of autonomy
identity insecurity
distrust
alienation
reduced willingness to take risks
conformity
cynicism,
and eventually disengagement
Kellogg et al. pointed out that this becomes particularly problematic when AI is not introduced and managed effectively, which, as mentioned earlier in this article, is often the case. In particular, opacity surrounding AI-driven decision-making, stemming from corporate secrecy, special technical expertise required to understand these systems, and the inherent opacity of machine learning itself are primary contributing factors.
This lack of transparency makes it difficult for employees to understand how management decisions are reached, what data inform these decisions, and/or how those decisions may affect employees. As a result, employees are asked to behave in a certain way without being provided with a clear understanding of the reasons underlying the decision-making process.
For instance:
If a human manager fires you, you can at least construct a story:
“My manager dislikes me.”
“My manager made a mistake.”
“I can explain what happened.”
“I can appeal to HR.”
With an algorithm, your analysis is will likely begin and end with:
“The system decided.”
And the system may not necessarily explain its reasoning.
That creates a particularly powerful form of powerlessness:
You aren’t merely being controlled; you don’t know what you need to do to stop being controlled.
Employees can essentially find themselves saying:
“I know I am capable of doing this work, but the system no longer recognizes me as someone who is allowed to do my job.”
That’s much more psychologically threatening than ordinary performance monitoring.
This transition toward algorithmic control made me think of several ideas in Hanna Arendt’s book The Origins of Totalitarianism³, especially her argument about statelessness, the “right to have rights,” factual reality, and the ability of political power to determine what counts as reality. Arendt argued that losing political/legal membership could mean losing the practical basis on which one’s rights were recognized.
And this connects surprisingly well with algorithmic control.
Kellogg et al.’s paper is fundamentally about organizations gaining the ability to observe, classify and act upon workers through data. Their six mechanisms are restricting, recommending, recording, rating, replacing, and rewarding.
Arendt’s concern takes this one step further.
Suppose your manager says:
“You’re late.”
You can respond:
“No, I wasn’t. I arrived at 8:58.”
There is a disagreement between two people, you and your manager.
But suppose an algorithm says:
“Employee arrival time: 9:17.”
And the organization treats the database as the authoritative record. Now you have an entirely different problem. You aren’t simply arguing about your behavior.
You’re arguing about the reality represented by the system.
And if you cannot access or modify the underlying data, you may not even know why the system reached its conclusion.
That is very Kafkaesque. In Franz Kafka’s book The Trial⁴, Josef K. is accused and subjected to a bureaucratic legal system, but he never fully understands the crime with which he is being accused, the rules that govern the process, and/or how he can meaningfully defend himself. The system’s authority does not depend on K. understanding or agreeing with it.
That can produce what we might call epistemic helplessness:
“I know something about myself, but I cannot make the institution recognize what I know.”
Seligman and Maier⁵ demonstrated in their 1967 paper that when individuals repeatedly experience situations in which their actions appear to have no influence over outcomes, they may eventually stop attempting to exercise control, even when opportunities to do so become available.
This connects directly to the problems Kellogg et al. identified around opaque algorithmic control: workers may not know why they were rated, why they lost opportunities, or how to change the system’s assessment of them.
Then comes the psychological problem of self-doubt. This is potentially even more powerful than the other factors. Imagine repeatedly being told by a system that:
your performance is poor
your behavior is abnormal
your productivity is inadequate
Eventually, the problem isn’t simply the external restriction. You may begin asking:
“What if the system is right?”
That is psychologically significant. The problem becomes particularly severe when algorithmic decisions affect a worker’s ability to participate economically.
Kellogg et al. noted that workers can “lose their jobs and wages, with no explanation and no opportunity to appeal.”
The psychological consequence is therefore not simply stress, but a loss of agency: Employees may be affected by decisions that determine their livelihood without being afforded any meaningful opportunity to understand or contest the basis of those decisions.
Based on this understanding, we arrive at perspective 7: Sense of Agency and Self.
Consider the following two opposing perspectives:
A) “The system can protect me from the subjectivity of human managers by evaluating my work according to consistent criteria, while giving me clear guidance on what is expected of me and how I can improve.”
B) “What happens when the organization’s criteria begin to define who I am supposed to be? If I have to continually adapt myself to an algorithm I cannot understand or question, I may gradually lose the sense that I am choosing who to be at work entirely—and with it, my agency and sense of self.”
Kellogg et al. emphasized that algorithmic control can make organizational control more comprehensive and rationalized while also generating resistance and “algoactivism.”
Employees don’t simply submit to management’s attempt to control them, they constantly engineer ways to preserve their autonomy. It’s essentially an arms race. Kellogg et al. outlined five key ways workers negotiate this control:
Avoiding the algorithm: Shielding specific tasks or data from the system’s visibility.
Adapting to the algorithm: Decoding the algorithm’s incentives and strategically adjusting behavior to match.
Gaming the algorithm: Manipulating inputs to force a more favorable output.
Collective action against the system: Organizing with peers to resist or reshape system rules.
Counter-use the algorithm: Exploiting the algorithm’s own logic and blind spots against itself.
Members of an organization are ideally active strategists rather than passive subjects of automated discipline. However, a critical tension arises regarding the loss of selfhood. While algoactivism seems like an assertion of agency, it carries a hidden cost. To resist effectively, workers must internalize the algorithm’s logic. True autonomy becomes compromised when even resistance requires shaping one’s behavior around the machine. This assumption gives rise to two final potentially powerful perspectives:
Perspective 8: Resist and Take Back Control.
A) “I can learn how the system works and use that knowledge to protect my autonomy, challenge its judgments, and find ways to maintain control over my work.”
B) “If I have to constantly adapt myself to what the system rewards, am I really exercising agency—or am I gradually becoming the kind of worker the algorithm wants me to be?”
And/or
A) “Even when I cannot challenge the system on my own, I can regain agency by connecting with other workers, sharing our experiences, and collectively demanding greater transparency, accountability, and control over how we are evaluated.”
B) “What if the very system controlling me is also monitoring who I talk to, what I say, and how I interact with my colleagues? When every conversation can become data and every interaction can potentially affect how I am evaluated, forming the collective relationships needed to resist the system may itself become risky—and eventually, I may not even know who I can trust.”
The most powerful form of algorithmic control for organizations may therefore be one that prevents resistance before it can become collective. If the system monitors not only what we do but who we speak to and what we say to one another (which already happens in some organizations), it can make solidarity itself a risk — leaving us isolated, unable to distinguish between a colleague and a potential source of surveillance, and ultimately unable to imagine how we might regain control together. And that’s a pretty scary perspective.
The same logic applies from the employees’ perspective: if organizations can use AI to prevent workers from organizing, then workers need to find strategies to prevent organizations from deploying AI in ways that undermine their ability to organize in the first place. The question is therefore no longer simply whether employees can resist algorithmic control, but whether they can retain the collective agency necessary to resist anything at all.
When you consider perspectives 7 and 8, which side do you come down on—A or B? And, more importantly, which perspective better protects us against AI solutions that are poorly introduced or misused as a tool of workplace control—and what might we risk if we get that choice wrong?
P.S. If you’re interested in understanding human interaction, perspective-taking, and mentalization more deeply — and learning how to apply these ideas in practice — I’m running a small-group workshop using the CAToM (Center for Applied Theory of Mind) Framework. Learn more → www.appliedtom.com/events
References:
Ferris, D. L., Lian, H., Brown, D. J., Pang, F. X. J., & Keeping, L. M. (2010). “Self-Esteem and Job Performance: The Moderating Role of Self-Esteem Contingencies.” Personnel Psychology, 63(3), 561–593.
Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366–410. https://doi.org/10.5465/annals.2018.0174
Kafka, F. (1998). The trial: A new translation, based on the restored text (B. Mitchell, Trans.). Schocken Books.
Seligman, M. E. P., & Maier, S. F. (1967). Failure to escape traumatic shock. Journal of Experimental Psychology, 74(1), 1–9. https://doi.org/10.1037/h0024514
