The Manager Is an Algorithm

Your boss might not be human anymore.

Across warehouses, gig platforms, call centers, and offices, algorithms now perform core management functions: scheduling shifts, assigning tasks, monitoring performance, and deciding who gets fired. This isn't automation of work — it's automation of management. The implications are profound.

When AI manages workers, it doesn't just change how decisions are made. It transforms the employment relationship itself. Workers become data points. Judgment calls become optimization problems. And the asymmetry of information and power between employer and employee expands dramatically.

This is not entirely new. As Langdon Winner argued in "Do Artifacts Have Politics?" (1980), technologies have always encoded power relations. The factory clock, the assembly line, the time-and-motion study — each was a management technology that reshaped worker autonomy. Algorithmic management is the latest and most comprehensive iteration of a centuries-old pattern. What is new is the scale, the granularity, and the opacity.

What Algorithmic Management Means

Traditional management involves humans making decisions about workers: who does what, when, how well, and whether they keep their job. Algorithmic management transfers these functions to software systems that collect data, process it, and take action — often without human review.

This is distinct from using software as a tool. A spreadsheet helps a manager track performance; algorithmic management is the manager. The difference matters because algorithms can operate at scale, in real time, and with a consistency that human managers cannot match — for better and worse.

Alexandra Mateescu and Aiha Nguyen of Data & Society define algorithmic management as "a diverse set of technological tools and techniques to remotely manage workforces, relying on data collection and surveillance of workers to enable automated or semi-automated decision-making." The key elements: remote control, continuous surveillance, and automated decisions.

The Six Mechanisms of Algorithmic Control

Katherine Kellogg, Melissa Valentine, and Angèle Christin's foundational 2020 paper "Algorithms at Work" identifies six mechanisms through which algorithms exert control over workers:

1. Directing

Algorithms assign tasks, routes, and schedules. Uber decides which driver gets which ride. Amazon's system tells warehouse workers which item to pick next and which path to walk. Instacart determines which orders go to which shoppers.

This sounds like efficiency. But directing also means workers lose discretion over their work. The algorithm optimizes for the company's metrics — speed, cost, utilization — not for worker preferences, safety, or wellbeing. A warehouse worker can't take the longer route to rest their back. A delivery driver can't skip the dangerous intersection.

2. Evaluating

Algorithms assess performance continuously. Ratings, metrics, scores — all computed in real time. Call center workers see their "handle time" and "sentiment scores" on screen. Uber drivers watch their ratings fluctuate with each ride. Amazon pickers are measured in "units per hour."

Evaluation by algorithm differs from human evaluation in crucial ways. It's constant, not periodic. It's quantified, not nuanced. It can't account for context: the customer who was always going to give one star, the package that was heavier than usual, the traffic that made the delivery late.

As Ifeoma Ajunwa documents, algorithmic evaluation often claims objectivity while encoding subjective choices. Someone decided which metrics matter, how to weight them, what thresholds trigger consequences. These are management decisions — they're just hidden inside the code.

3. Disciplining

Low scores have consequences. Uber drivers below 4.6 stars risk deactivation. Amazon workers who miss "rate" targets receive automated warnings, then termination. Content moderators who fall below accuracy thresholds lose access to tasks.

Algorithmic discipline is swift and often automated. There's no appeal to a manager who might understand the circumstances. The algorithm doesn't know your child was sick, your car broke down, or the customer was abusive. It knows your numbers.

This creates what Veena Dubal calls "algorithmic wage discrimination" — a system where workers' actual compensation varies based on opaque algorithmic decisions about which tasks they receive, how those tasks are priced, and whether their account remains active.

4. Recommending

Some algorithmic management operates through suggestion rather than command. The system recommends a schedule, proposes a route, suggests which tasks to prioritize. In theory, workers retain choice.

In practice, recommendations are often coercive. Declining an Uber ride recommendation hurts your acceptance rate. Ignoring the suggested route may count against you if the delivery is late. The "recommendation" is a command dressed in softer language.

5. Restricting

Algorithms limit what workers can see and do. Uber drivers don't know their destination until they accept the ride. Gig workers can't see all available tasks, only those the algorithm chooses to show them. Workers are restricted from understanding the very system that governs their work.

Alex Rosenblat's Uberland documents how Uber's information asymmetry functions as control. Drivers can't make informed decisions about which rides to accept because they lack basic information. They can't coordinate with other drivers because the platform intermediates all communication. The restriction of information becomes a mechanism of power.

6. Recording

Everything is tracked. GPS location, keystrokes, time between actions, facial expressions, tone of voice. Algorithmic management depends on continuous data collection — surveillance is the foundation.

This recording creates what some researchers call a "digital panopticon." Workers must assume they're always being watched, always being measured. The psychological effect extends beyond the data actually collected. The possibility of surveillance shapes behavior.

The Paradigm Case: Amazon Warehouses

Amazon's fulfillment centers represent the most developed form of algorithmic management. Understanding how they work illuminates the broader pattern.

The technical system:

Workers carry handheld scanners that track their location and the items they handle. The system assigns tasks — go to this location, pick this item, place it here — with precise timing expectations. "Rate" is calculated in real time: items per hour.

The system knows where every worker is, what they're doing, and how fast they're doing it. Time off task — TOT — is tracked in minutes. Taking too long between scans triggers flags. The algorithm doesn't distinguish between a bathroom break, a conversation with a coworker, or genuine slacking.

The consequences:

Emily Guendelsberger's On the Clock documents the experience from inside. Workers describe feeling like "robots" — not because they do mechanical work, but because they're controlled by a mechanical system. There's no room for judgment, no space for the informal negotiations that make work bearable.

Injury rates at Amazon warehouses are notably higher than industry averages. A 2023 investigation found Amazon warehouses had injury rates nearly double those of non-Amazon warehouses in the same industry. Workers and safety advocates attribute this partly to algorithmic pressure — the system demands speed that bodies can't safely sustain.

The termination machine:

Amazon's system can generate termination decisions automatically. Workers who fall below rate thresholds receive warnings — also automated. The process can proceed from data collection to firing with minimal human involvement.

In 2019, reporting revealed that Amazon had fired hundreds of workers at a single facility for productivity reasons. The company defended the practice as "performance management," but the scale and automation raised questions about whether meaningful human review was occurring.

Gig Economy: Management by App

Ride-hailing and delivery platforms extend algorithmic management to workers classified as "independent contractors." This combination — algorithmic control without employment protections — represents a distinctive labor arrangement.

Uber and Lyft:

Rosenblat's research documented how Uber maintains control while denying employment. Drivers are "independent" but can't set their own prices, choose their customers, or work without the app's mediation. The algorithm directs, evaluates, and disciplines — all functions of a manager — while the legal structure says there's no employment relationship.

Dynamic pricing (surge) and ride assignment create powerful incentives. Drivers learn that certain behaviors (accepting more rides, working at certain times, maintaining high ratings) lead to better outcomes. The algorithm shapes behavior through consequences, even without explicit commands.

Food delivery:

DoorDash, Instacart, and similar platforms add complexity. Workers shop for items, handle substitutions, navigate stores — tasks requiring judgment. But the algorithm evaluates them on speed metrics that don't account for this complexity.

Workers report gaming the system: memorizing store layouts, pre-shopping orders, developing shortcuts. The human worker adapts to the algorithm's expectations, even when those expectations don't match the work's reality.

The information gap:

Gig workers consistently report that they don't understand how the algorithm works. Why did I get this ride and not that one? Why did my earnings drop this week? What determines my "priority" status?

This opacity is functional. If workers understood the algorithm, they could optimize against it. Platforms keep the system opaque to maintain control. But this creates a profoundly unequal relationship: the employer sees everything about the worker; the worker sees almost nothing about the employer's decision-making.

Call Centers: Sentiment Under Surveillance

Call center work has long involved monitoring. But AI has transformed the scale and nature of surveillance.

Real-time monitoring:

Modern call center systems use AI to analyze calls as they happen. Voice analysis detects "sentiment" — is the customer angry? Is the worker sounding frustrated? The system can prompt workers in real time: slow down, speak more positively, use the customer's name.

This creates a kind of emotional management. It's not enough to do the work; you must perform the correct emotions while doing it. Arlie Hochschild called this "emotional labor" decades before AI made it measurable.

The metrics:

Average handle time, first call resolution, customer satisfaction scores, script adherence. These metrics drive algorithmic evaluation. Workers who hit targets get better shifts and performance ratings. Those who don't face consequences.

But the metrics often conflict. Resolving a complex problem takes time, which hurts average handle time. Genuine help may require deviating from the script. Workers must balance customer needs against algorithmic demands.

Management response:

Some call centers have deployed AI to manage the managers. Systems track supervisors' coaching activities, measure their teams' performance, recommend interventions. The surveillance extends up the hierarchy, though rarely to executives.

Content Moderation: Trauma as Data

Content moderators — the workers who review disturbing material so users don't have to see it — face a distinctive form of algorithmic management. Their exposure to trauma is governed by algorithmic queues and quotas.

The work:

Moderators review content flagged by users or AI systems. They decide: does this violate platform rules? The content includes violence, sexual abuse, hate speech, self-harm — the worst of what humans post online.

Sarah Roberts' Behind the Screen documents the conditions. Workers often have seconds to make decisions. They're evaluated on speed and accuracy. The system assigns content without regard to its cumulative psychological impact.

Algorithmic queues:

The queue is endless. Finish one piece of content, another appears. There's no natural break, no endpoint. The algorithm determines pace, and the pace is relentless.

Workers report that requesting breaks hurts their metrics. The system doesn't distinguish between needing time to recover from disturbing content and simply being slow. Human need confronts algorithmic indifference.

Who does this work:

Like other forms of hidden AI labor, content moderation concentrates in the Global South. The Philippines, Kenya, India — workers earning a fraction of Silicon Valley wages while protecting American users from harmful content. The redistribution runs the wrong direction: trauma exported, profits retained.

The "Anti-Bias" Claim

Proponents argue algorithmic management reduces discrimination. Algorithms don't have racial prejudices, gender biases, or personal grudges. They just process data.

Ifeoma Ajunwa's "The Paradox of Automation as Anti-Bias Intervention" dissects this claim. The argument fails for several reasons:

Algorithms encode existing biases:

Training data reflects historical discrimination. If past promotion decisions favored certain groups, an algorithm trained on that data will reproduce the pattern. The bias isn't eliminated — it's laundered through technology that claims objectivity.

Proxy discrimination:

Algorithms can discriminate without using protected categories. Zip code, commute time, social media activity — facially neutral factors that correlate with race, class, gender. The effect is discriminatory even if the mechanism isn't explicitly so.

Opacity hides discrimination:

When a human manager discriminates, there's at least the possibility of identifying the decision-maker and holding them accountable. When an algorithm discriminates, the causal chain is obscure. Who decided which variables to include? Who set the weights? Who chose the training data? Accountability diffuses into the system.

Feedback loops:

Algorithmic decisions affect future data. If workers in certain neighborhoods receive fewer assignments, their metrics suffer, justifying fewer assignments. The algorithm creates the data that confirms its decisions. Past discrimination becomes self-perpetuating.

What Workers Experience

Research consistently documents the psychological impact of algorithmic management:

Loss of autonomy:

Workers report feeling like extensions of the machine rather than independent actors. The algorithm decides what to do and when to do it. Judgment, discretion, the small adaptations that make work bearable — all constrained.

Constant surveillance:

The knowledge of being watched creates stress independent of actual consequences. Workers modify behavior preemptively. Bathroom breaks become strategic decisions. Conversations with coworkers feel risky.

Precarity:

Algorithmic management often accompanies precarious employment: gig work, contractor status, at-will employment. The combination intensifies vulnerability. Not only is the system opaque, but challenging it risks losing the job entirely.

Depersonalization:

There's no one to appeal to. The algorithm can't exercise discretion, consider context, or make exceptions. Workers describe feeling that no one is in charge — or rather, that the one in charge can't be reasoned with.

Gaming the system:

Workers develop strategies to optimize for algorithmic metrics. Uber drivers learn which rides to avoid, which times to work. Amazon workers find ways to boost "rate" that may not actually improve productivity. The gap between algorithmic measurement and actual work quality becomes a space for adaptation — and cynicism.

Resistance: How Workers Push Back

Workers aren't passive recipients of algorithmic control. Resistance takes multiple forms:

Organizing:

The Alphabet Workers Union, Amazon Labor Union, and campaigns among gig workers represent emerging efforts to build collective power. Algorithmic management is a frequent grievance — workers demand transparency, appeal rights, and limits on surveillance.

The Amazon Labor Union's success at Staten Island's JFK8 facility in 2022 was notable partly because it happened despite algorithmic obstacles to organizing. The system makes traditional tactics difficult: workers are dispersed, schedules vary, the "employer" is technically a platform.

Litigation:

Workers and advocates have challenged algorithmic management through courts and regulatory agencies. Cases have targeted:

• Worker classification (are gig workers employees?)
• Discrimination in algorithmic hiring and evaluation
• Wage theft through algorithmic payment calculations
• Lack of due process in algorithmic termination

The outcomes are mixed. Some cases have produced settlements or regulatory guidance. Others have failed on procedural grounds or narrow interpretations of existing law.

Collective action:

Gig workers have organized strikes and work stoppages despite lacking union recognition. Drivers have coordinated surge pricing by going offline simultaneously. Delivery workers have protested tip policies. These actions demonstrate collective power even in the absence of formal structures.

Transparency demands:

Workers increasingly demand to understand the algorithms that govern them. What data is collected? How are decisions made? What factors affect ratings, assignments, termination? The right to understand algorithmic decisions has become a labor demand.

Policy Responses

Governments are beginning to respond to algorithmic management, though responses vary widely:

EU Platform Work Directive (2024):

The directive creates a presumption of employment for platform workers when certain conditions are met — reversing the legal burden that has let platforms classify workers as contractors. It also requires transparency about algorithmic decision-making and creates rights to human review of significant automated decisions.

This is the most comprehensive response to date. But implementation varies by member state, and platform companies are adapting business models to evade classification triggers.

California AB5 and Proposition 22:

California's AB5 (2019) attempted to classify gig workers as employees by default. Uber and Lyft spent over $200 million on Proposition 22 (2020), a ballot measure exempting them. The result: gig workers in California remain contractors but with some limited benefits.

The episode demonstrated the political economy of algorithmic management. Platforms have resources to fight classification that workers lack to defend it. The full story is detailed in AI Gig Economy.

Right to explanation:

The EU's General Data Protection Regulation (GDPR) includes rights related to automated decision-making. Workers can request information about the logic of algorithmic decisions affecting them. But enforcement has been limited, and "explanation" often means simplified, unhelpful descriptions.

US regulatory approaches:

The Federal Trade Commission has indicated interest in algorithmic management practices, particularly surveillance and deceptive representations of gig work. The National Labor Relations Board has examined whether algorithmic surveillance chills organizing. The Department of Labor has issued guidance on worker classification.

But comprehensive federal legislation remains elusive. The patchwork of state laws and agency guidance leaves workers with uneven protections.

Emerging proposals:

Policy advocates have proposed various interventions:

• Algorithmic accountability acts requiring impact assessments
• Portable benefits systems decoupled from employment status
• Data rights giving workers ownership of their performance data
• Sectoral bargaining allowing collective negotiation across platforms

The Redistribution Question

Algorithmic management is, at its core, a redistribution story. It shifts power, information, and value from workers to employers.

Power redistribution:

Management decisions that once required human judgment now happen automatically. This isn't delegation to subordinates; it's concentration in systems that employers control entirely. Workers have no seat at the table because there is no table — only code running on servers.

Information redistribution:

Employers know everything about workers: location, speed, ratings, patterns. Workers know almost nothing about how decisions are made. This asymmetry is designed, not accidental. Information is power, and algorithmic management concentrates it.

Value redistribution:

The "efficiency gains" from algorithmic management come from somewhere. Workers report intensified pace, reduced breaks, constant pressure. The productivity increases extract value from workers — their time, their bodies, their attention. Profits flow to shareholders; costs land on workers.

Risk redistribution:

Algorithmic management often accompanies employment structures that shift risk to workers. Gig workers bear the costs of slow periods, bad ratings, and platform changes. Contractors lack unemployment insurance, workers' compensation, and job security. The efficiency of algorithmic control combined with the flexibility of precarious employment creates maximum vulnerability.

What would redistribution in the other direction look like? Workers with access to the data collected about them. Transparency about how decisions are made. Due process before significant automated decisions. Collective bargaining over algorithmic design. Shared governance of the systems that govern work.

This isn't futurism. These are features that could be built into current systems — if the people who control them chose to build them, or were required to by law.

Some employers are already moving in this direction — not out of altruism, but because they are finding that transparency and worker input improve system design. Companies that involve workers in piloting algorithmic tools report higher adoption rates, fewer errors, and less turnover. The collective bargaining agreements emerging in Europe, particularly under Germany's co-determination system, demonstrate that algorithmic management can be governed through negotiation rather than unilateral imposition. The question is whether competitive pressure will drive a race to the bottom or whether democratic governance frameworks and the EU AI Act will establish floors that prevent the worst practices.

What Comes Next

Algorithmic management is expanding. AI capabilities are advancing. The same systems deployed in warehouses and gig platforms are coming to white-collar work: AI monitoring of emails, automated performance assessment, algorithmic scheduling for professional workers.

The choices being made now will shape working conditions for decades. Whether algorithmic management becomes a tool for extraction or augmentation depends on who has power over its design and deployment.

Workers asking "Will AI take my job?" might also ask: "Is AI already my boss?" For millions, the answer is yes.

Related

• Hidden Labor
• Automation Debate
• Ghost Work
• AI Gig Economy
• Content Moderation Mental Health
• Collective Bargaining
• White-Collar Displacement
• Automating Inequality
• Why Removing Bias Doesn't Work
• EU AI Act Takes Effect
• Democratic AI Governance
• Regulatory Capture

Sources

• Ajunwa, Ifeoma. "The Paradox of Automation as Anti-Bias Intervention." Cardozo Law Review 41, no. 5 (2020). SSRN
• Dubal, Veena. "Algorithmic Wage Discrimination." Columbia Law Review 123, no. 7 (2023). Columbia
• Guendelsberger, Emily. On the Clock: What Low-Wage Work Did to Me and How It Drives America Insane. Little, Brown, 2019. Publisher
• Kellogg, Katherine C., Melissa A. Valentine, and Angèle Christin. "Algorithms at Work: The New Contested Terrain of Control." Academy of Management Annals 14, no. 1 (2020). AoM
• Mateescu, Alexandra, and Aiha Nguyen. "Explainer: Algorithmic Management in the Workplace." Data & Society, 2019. Data & Society
• Roberts, Sarah T. Behind the Screen: Content Moderation in the Shadows of Social Media. Yale University Press, 2019. Publisher
• Rosenblat, Alex. Uberland: How Algorithms Are Rewriting the Rules of Work. University of California Press, 2018. Publisher
• Strategic Organizing Center. "The Injury Machine: How Amazon's Production System Hurts Workers." 2023. SOC
• Winner, Langdon. "Do Artifacts Have Politics?" Daedalus 109, no. 1 (1980). JSTOR