In September 2020 a Vice reporter discovered that Amazon was seeking to hire two “intelligence analysts” into its Global Security Operations division (GSO). The analysts would use data analytics and other tools to detect and resist “labor organizing threats” and other political opposition to the company. Later that month an Amazon employee alleged that GSO had also monitored internal company message boards to spot union organizing, focusing on boards developed by workers of color and other groups typically underrepresented in Silicon Valley. Such pervasive worker surveillance backfired on Amazon in at least one case. Employees at an Amazon warehouse on Staten Island unionized in 2022, in part out of frustration over incessant automated productivity monitoring. Corporate surveillance has also contributed to organizing among Uber and Lyft drivers, Apple store workers, and employees at Tesla’s self-driving program.

Surveillance technologies materialize today’s workplace class politics.

These efforts have implications for longstanding debates around technology and the future of work. For more than a decade, scholars, journalists, and tech leaders have focused on two ways that data-driven technologies are altering jobs: by automating tasks and therefore displacing certain workers, and by discriminating on the basis of race, sex, national origin, or disability. Those are critical issues, but surveillance technologies are having another effect on work as well. Companies across today’s vast service economy are using such technologies as tools of class domination, deploying them to limit wage growth, prevent workers from organizing, and enhance labor exploitation. Workers’ increasing resistance to surveillance is therefore also a process of class formation—and reforms that support such resistance could encourage a more democratic politics of workplace technology.

To be sure, these conflicts are not new. For well over a century, companies have sought to generate, capture, and quantify information about workers and work processes and use it to suppress wages. The transition to heavy industry involved wresting control of production away from craft workers and their unions, who controlled apprenticeship programs, chose technologies, and established output rates and wages as the so-called “law” for their crafts. Craft workers were aware that industrial engineers were seeking to replicate their skills and would often resist monitoring, or even refuse to work at all if supervisors were present.

Employers won that battle and have deployed information as a power resource ever since. Often that involves using technology to make work more legible to a central authority, and therefore more susceptible to centralized control. The telegraph, telephone, fax machine, and modern information technologies all enable the integration of enterprises across vast distances, generating economies of scale but also facilitating supervision of far-flung workforces. Companies’ surveillance capacities have nevertheless expanded dramatically in recent decades with the maturation of corporate intranets, mobile computing, location tracking, image and natural language recognition, and other forms of advanced data analytics. Today companies often aspire to constant surveillance that reaches all aspects of work and production. Surveillance today is also asymmetric: companies can monitor workers without them knowing while simultaneously preventing workers from monitoring management. In that sense, surveillance technologies materialize today’s workplace class politics.

Amazon’s warehouses exemplify this new regime, as is apparent in documents that were disclosed as part of a labor dispute between the company and a worker who alleged that he had been fired in retaliation for union organizing. They stated that “Amazon . . . tracks the rates of each individual associate’s productivity,” and that the company “automatically generates any warnings or terminations regarding quality or productivity without input from supervisors.” Around 300 workers in that warehouse, over 10 percent of the staff, were terminated via that process in a 12-month period.

Economic and legal shifts over the last several decades have led companies to use novel technologies in this manner. In the 1970s oil shocks, competition in global manufacturing, stagflation, and the costs of maintaining an imperial military generated a profitability crisis. Our policy responses first encouraged deindustrialization—which eviscerated organized labor—and then encouraged the growth of today’s service economy, where our largest employers are in retail, food service, logistics, hospitality, and healthcare. Many such firms employ massive numbers of workers but suffer from low productivity growth because their product requires human labor or attentiveness that is difficult to augment through technology. As a result they have especially strong incentives to suppress wage growth. Many embrace a business model with high employment levels but low skills and high turnover, and use new technologies to prevent workers from building collective power.

Our largest employers are in retail, food service, logistics, hospitality, and health care.

The legal shifts have been more subtle but no less momentous. Employers have long enjoyed rights to monitor the performance of work in most cases. But after the 1970s crisis, employers—themselves under pressure from finance capital—pushed for greater freedom to hire and fire workers at will, to avoid unionization, to purchase labor while denying workers basic employment protections, and to treat the enterprise more like their own sovereign property. That left workers less able to resist ever-expanding surveillance. Indeed, workers are also vulnerable to off-the-job surveillance in many states, with employers checking their social media posts and monitoring their political activities without legal consequence.

To be sure, service companies also use novel technologies to enhance productivity where possible. Walmart set the template here in the 1980s and ’90s. The company used data on store-level inventory and the movement of goods around the globe to optimize its sourcing and distribution systems from its headquarters in Bentonville, growing into a retail behemoth in the process. But Walmart also pressed suppliers for ever lower costs, even driving down wages among those suppliers’ workers. FedEx similarly grew into a delivery giant by developing sophisticated logistics systems and by treating drivers as independent contractors, thus denying them basic labor law protections. Amazon built on Walmart’s model, combining global sourcing with an online sales platform. Uber and other gig economy companies built on FedEx’s model, developing matching algorithms for short-term logistical tasks to be performed by drivers without labor protections. In both cases data-driven technologies are used to make business processes more efficient and to ensure labor discipline.

Companies use data to discipline labor in three different ways. I call the first “digital Taylorism,” referring to early twentieth-century Taylorism, the system of scientific management that established managerial control over the labor process. Like its forebear, digital Taylorism involves intertwined forms of automation and intensified surveillance. Where possible, companies use algorithms and robots to perform tasks once performed by line-level workers, though not in the anthropomorphic manner envisioned by Hollywood writers. Rather, they break jobs into simpler tasks, some of which are automated while others are performed by workers with little specialized training. Then, regardless of which tasks workers are performing, companies use new algorithmic means to assign tasks and to schedule and oversee workers.

Consider a “picker” in one of Amazon’s semi-automated warehouses, whose job consists of grabbing items off shelves brought to them by robots. The robots and shelves are physically separated from workers by a reinforced chain-link fence, which is necessary to minimize danger as the robots have limited abilities to sense their environment. The robots move quickly and constantly, guided by barcodes on the floor, often driving right toward one another and then pivoting at the last moment and moving ninety degrees in another direction. Their movements are directed and optimized by suites of algorithms, with the overarching goal of minimizing the time to fill orders, and reflecting a sort of alien intelligence.

Companies face no requirements to clear novel information technologies before deploying them in consumer markets or workplaces.

The human workers, however, are little more than production inputs. Each picker works in a phone booth-sized area, where they often cannot make eye contact with others and where they are constantly monitored by video cameras and image-recognition algorithms. A large digital stopwatch shows how long they take to perform each task, down to the second. Once a picker has filled a bin with goods, a network of automated conveyor belts sends it to be packed and shipped by workers who are similarly surveilled. Vending machines just off the work floor supply ear protection, gloves, and the Advil that many workers take regularly to alleviate pain and inflammation from repetitive stress. Algorithmic monitoring systems may report workers who do not perform quickly enough or who might take a bathroom break without a manager’s clearance, sometimes even recommending termination. Given the simplicity of these jobs, Amazon can plug workers in and out of them easily and risks little by terminating those who protest.

Companies can also use data gathered from workers to replicate and then alienate their skills. For example, Uber has integrated GPS–powered navigation into the driver side of its app—and aims to improve it continuously using data from past rides. But taxi drivers’ specialized knowledge of how to navigate a crowded city historically gave them some labor market power. In London cab drivers must even pass a test demonstrating knowledge of the names and locations of all streets in the area so that they could get to any location without a map. Uber has therefore captured some of taxi drivers’ tacit knowledge, claimed legal property rights to it, and leased that property back to drivers.

In addition to digital Taylorism, companies use surveillance and data aggregation techniques to prevent unionization and other forms of collective action. For example, companies may be able to use new recruiting algorithms to aggregate data on applicants’ employment history with data on their social media posts or political behavior, and then screen out workers who are likely to challenge management’s authority. They may also be able to deter organizing simply through pervasive surveillance. Worker organizing is often a process of building trust and solidarity. When workers are invested in a shared identity, they can take collective action to protect one another—and may well prevail despite management’s near-certain opposition. But modern surveillance can prevent such mobilization. For one thing, workers who are constantly supervised and physically separated from one another have little opportunity to make common cause. Moreover, the growth of speech processing and natural language processing software could make it possible for companies to “hear” nearly everything said in a workplace and see when workers are speaking with one another, which could facilitate crackdowns on union talk—or indeed, any kind of talk.

Companies can also use new surveillance devices—and their own legal entitlements to control the workplace—to suppress unionization more directly. Some tech giants have been developing technically sophisticated but politically crude union avoidance tools. In an internal presentation in June 2020 regarding “Facebook Workplace,” a chat and collaboration platform meant to compete with Slack, Facebook executives noted that administrators could remove posts on certain topics and prevent them from trending. Among the terms moderators suggested companies might want to block was “unionize.” That sort of effort, and Amazon’s parallel efforts to detect “labor organizing threats,” likely run afoul of the law—and the NLRB’s General Counsel has signaled a desire to clamp down on such practices. But given the difficulty of detecting those efforts in the first place, and the NLRB’s limited resources, even clear union-busting may go undeterred in many instances.

Workers are also vulnerable to off-the-job surveillance, with employers checking their social media posts and monitoring their political activities without legal consequence.

Finally, companies are using new technologies to alter the scope of their enterprises. They are purchasing labor without hiring workers as legal employees, treating them as independent contractors, for example, or using franchising arrangements. These strategies are not new, but many companies today employ them even as they surveil and manage those workers as closely as they would traditional employees. Companies are also exploiting advanced information technologies to build substantial market power. Today that often involves a new business form, the internet-mediated platform. Uber and Amazon, for example, both connect outside parties to exchange goods and services. As technology scholars have argued, successful platforms often grow rapidly and come to dominate their sectors due to network effects and economies of scale. As a result, power in many low-wage sectors today is highly concentrated in a few corporate headquarters, but legal responsibility for working conditions is diffuse.

McDonald’s, for example, is not a single legal enterprise, but an amalgamation of tens of thousands of formally distinct entities with McDonald’s corporate at the center. The company runs some locations itself; others are McDonald’s franchises, independently owned and operated as separate companies. But workers and the NLRB have alleged that McDonald’s uses point-of-sale devices and other workplace technologies to track sales and labor costs, to set schedules, and even to track how long workers take to complete orders. Amazon, similarly, has outsourced delivery to various outside companies it terms “Delivery Service Partners” (DSPs). As one journalist found, Amazon’s contracts require DSPs to “provide Amazon physical access to their premises and all sorts of data the retailer wants, such as geo-locations, speed and movement of drivers—information the company says it has the power to use however it wants.” Such monitoring efforts can give Amazon the powers traditionally associated with employment without the duties and costs.

All of these trends culminate in today’s labor politics, where knowledge and control are centralized, and line-level workers have little autonomy and no voice. Even when laboring side by side, workers stand at the end of surveillance spokes that extend from a corporate nerve center. They then must compete with one another for jobs and desired shifts and therefore face substantial market discipline. As a result, service workers are increasingly a class in a structural sense, even if they don’t always take class-based action.

What would it mean to democratize the governance of workplace data and technology? A first step would be to make it easier for workers to build collective power. Labor’s current legislative proposal, the Protecting the Right to Organize Act (PRO Act), would help the NLRB deter and remedy unfair labor practices, streamline the union certification process, and expand workers’ rights to strike. In recent years scholars and some unions have also suggested more ambitious reforms to our labor law: granting collective bargaining rights to unions that have substantial but not majority support in a workplace; making it easier to bargain with all the employers in a given sector at once; or even establishing forms of collective workplace representation as a matter of right, so workers do not have to fight for it.

Service workers are increasingly a class in a structural sense, even if they don’t always take class-based action.

But even those efforts would not fully address employers’ ability to use data to demobilize workers. That may require a more fundamental shift in our labor law: eroding employers’ longstanding authority to choose workplace technologies and giving workers a real voice in production planning and execution. Three categories of reforms to data practices can advance those goals: bans on data collection and usage in various instances, subjecting data practices to bargaining in other cases, and placing other data sources or technologies under public or social control. In other words, abolishing, bargaining, and socializing data.

Workplace data abolition builds on social movements’ demands to de-digitize various spheres of social life. As Ben Tarnoff argues, some technologies—such as predictive policing and police-controlled facial recognition—mainly exist to enact “relationships of domination.” Dismantling those technologies can therefore create space to develop new and more democratic social relations. The notion that certain technologies should be simply abolished challenges much technology policy and ideology in the United States. Unlike in the case of drug regulation, for example, companies face no requirements to clear novel information technologies before deploying them in consumer markets or workplaces. And yet legislators have often banned certain forms of information and data gathering in the workplace: states have long regulated drug testing, for example, and many states have also prohibited employers from requiring employees and applicants to provide their social media passwords.

Given employers’ extraordinary technological capacities today, policymakers may need to consider bans on forms of workplace surveillance that are long established and seem uncontroversial, such as the monitoring of workers on the shop floor as they perform work tasks. Indeed, advocates have begun discussing approaches to workplace data that involve a degree of abolition. Researchers at the University of California Berkeley Labor Center, for example, developed a set of recommendations around workers’ technology rights following broad consultation with scholars, unions, and others. Their report proposed a ban on worksite facial recognition and the use of algorithms to try to discern workers’ emotions, as well as restrictions on employers’ collection of worker data that is not “necessary and essential for workers to do their jobs.” That same report proposed that employers use electronic surveillance only when “strictly necessary to enable core business tasks, to protect the safety of workers, or when needed to comply with legal obligations.”

Bargaining mandates would overlap with and supplement abolition efforts. In fact, they may be preferable in cases where companies claim the technology at issue may enhance productivity. Requiring bargaining could then encourage companies to design or use those technologies in ways that do not harm workers. Bargaining can also enhance individual and collective self-governance. Novel information technologies today embed class relations, often in ways that are obscure to line-level workers. As noted above, Amazon warehouse workers and Uber drivers both report being disciplined for failing to meet performance standards they did not know existed or based on surveillance of which they were not aware. Requiring employers to make such technologies transparent and bargain over their implementation is then essential to preventing workers from being governed in arbitrary and unforeseeable ways.

By placing data sources under public or social control, we could seek a politics of technology embedded in a different set of class relations.

In the union sector, Congress could implement bargaining mandates by specifying that all uses of workplace technology are a “mandatory subject of bargaining” in labor law parlance. Congress could also consider giving workers some collective rights to consult on technological changes regardless of their unionization status, like the rights German workers enjoy through works councils. Workers’ optimal bargaining strategy when an employer seeks to implement a new technology would vary based on the circumstances. Amazon warehouse workers might welcome the integration of new robotic systems, for example, and fast-food workers might welcome tablet-based ordering systems, if the companies share productivity gains through higher wages or a more reasonable pace of work.

The final set of reforms would give workers and the public greater control over data and related technologies: socializing them in the sense of treating them like a public resource. For example, Congress could require companies to share much more of the data they gather on workers and work processes. Regulators or worker rights organizations could then analyze that data to spot basic labor law violations like wage and hour noncompliance. Regulators could also use that data to map companies’ power over workers and to hold them to appropriate duties. For example, evidence that Uber or McDonald’s monitored drivers or franchisees’ workers could be presumptive evidence of employment status, so that the companies would owe those workers minimum wages, workers’ compensation, and other protections.

Congress or the NLRB could also expand workers’ rights to access their employers’ proprietary technologies and data sources to bolster organizing efforts. Workers already use employers’ apps to organize in some cases. Gig economy workers, for example, have at times collectively turned off their apps to protest companies’ policies. Such protests could be more effective if the workers and organizers could use the apps themselves to contact and mobilize coworkers. Many service workers would also benefit from being able to communicate directly with customers via their companies’ apps or websites. This would be a digital analog of the in-person picket line, where in the past workers would solicit potential customers’ support outside a struck business. Finally, states and cities could do much more to encourage worker-owned cooperatives and to help them compete with for-profit businesses by helping them obtain funding and giving them preferential access to government contracts.

These reforms would not ensure decent work or social equality on their own. They would need to be coupled with massive state investments in care and social reproduction, for example, as well as in a green transition. But they would advance those goals by rendering technology and its governance more democratic. That project faces many political challenges, including an ingrained sense that technology is simply too complex to be a democratic medium. Sometimes that involves an experience of the sublime: countless movie plots now turn on breakthrough inventions that generate grand existential conflicts, and there is something both fascinating and terrifying about automated warehouses and the new crop of chatbots. In other cases it involves the mundane. As workers we accommodate the class relations materialized in technologies: driving a certain way, talking a certain way, following a clock. Neither experience—of the sublime or the mundane—entails a sense of individual or collective agency.

But that is an illusion. Technology is a product of our social knowledge and past labor, and it holds out the promise of basic material security and freedom from drudgery for all. We chafe against technologically mediated orders because we are social beings who need respect, community, and space for creativity. Those instincts could be the basis for a new politics of technology, one embedded in a different set of class relations, which ultimately seeks to subject production to real democratic control. Data-driven technologies could then help meet social needs and enable us to choose our labors, making them a source of dignity and self-worth.

Editors’ Note: This essay is adapted from Data and Democracy at Work: Advanced Information Technologies, Labor Law, and the New Working Class, just published by the MIT Press.