What Have We Learned? Who Is Watching the Work?

In January 1914, Henry Ford announced that workers at his Highland Park plant would earn five dollars for an eight-hour day, more than double the $2.34 they had been paid for nine hours. Ford had a reason to be generous. Workers disliked the new assembly line so much that turnover in 1913 ran at 380 percent. Crowds of applicants gathered outside the plant after the announcement.

The fine print mattered. Much of the raise was paid as "profit sharing," and only to employees the company judged worthy of it. To decide who qualified, Ford's new Sociological Department sent investigators on unannounced visits to workers' homes. They rated the cleanliness of kitchens, noted whether the family took in boarders, checked bank deposits, confirmed that children were attending school, and watched for divorce filings. A company pamphlet told employees it was their duty to help the investigators. Most workers accepted the arrangement because the wage was too good to refuse. Ford’s paternalism also had a much darker side: he became one of the country’s most prominent promoters of antisemitic conspiracy theories, using his newspaper, The Dearborn Independent, to publish material that was later translated and circulated in Germany and admired by Nazi leaders.

This month we look at the part of the employer-worker compact that governs how closely an employer may watch the people it pays. Employers have a legitimate interest in knowing whether work gets done, whether it is done safely, and whether company property and data are protected. Workers have an equally old interest in dignity, privacy, and not being treated as suspects. The tools have changed from clipboards to keystroke loggers and AI productivity scores, but Ford's bargain (better pay in exchange for being watched) is still recognizable. The governance question is whether what we measure actually helps the business create value, or merely makes work more visible. 

The man with a stopwatch

This month's "person with a mission" saw himself as a reformer. Frederick Winslow Taylor believed that most conflict between bosses and workers came from guesswork: nobody knew how long a job should take, so managers pushed and workers held back. His answer was measurement. Through time-and-motion studies, he broke jobs into their smallest steps, timed each one, and set a standard for how much output a worker could produce in a day. In Taylor's view, higher output would fund higher pay, and both sides would win. His Principles of Scientific Management (1911) became one of the most influential management books ever written.

The stopwatch also gave management a new kind of visibility. For the first time, the pace of every task could be observed, recorded, and compared against a standard set by someone who did not do the work. Workers understood what that meant. In 1911, Taylor's associates brought time study to the Army's Watertown Arsenal outside Boston, and the molders walked off the job. Their complaint, in a letter to the commanding officer, was that experienced craftsmen were now being told by a young man with a stopwatch that they were not working hard enough.

The strike led to House hearings, and Taylor was called to defend his system before Congress. The outcome surprised him. Congress attached riders to appropriations bills that banned stopwatch time studies in military and other federal workplaces, and the ban held for about three decades. When the Defense Department revived the methods after World War II, it applied them first to a new civilian workforce of women, minorities, and foreign nationals, the groups with the least power to object.

Two details from Watertown deserve attention from anyone rolling out monitoring or AI tools today. First, Taylor himself had warned the arsenal's officers not to introduce time study without preparing the workers first; they wanted quick results and skipped that step. The backlash came less from the measurement than from how it was imposed. Second, the workers' objection was about judgment as much as speed. They were not refusing to be productive. They were refusing to let someone who had never done their work decide what productive meant. That distinction still matters: activity is not the same thing as performance, and performance is not necessarily the same thing as value. 

Spies, lie detectors, and the lines Congress drew

Taylor's stopwatch was in plain sight. The surveillance that shocked Congress two decades later was hidden. In the 1930s, employers hired private detective agencies to place informants inside unions and report on who was organizing. From 1936 to 1941, a Senate subcommittee chaired by Robert La Follette Jr. of Wisconsin investigated the five largest agencies, including Pinkerton and Burns. The committee's reports listed about 2,500 firms that had spent more than $10 million on labor spies, guns, and tear gas, and found that General Motors alone paid close to $1 million for spy services in the mid-1930s. Robert Pinkerton, grandson of the agency's founder, defended the business on the grounds that an owner is entitled to stay informed about how his company is run.

La Follette's name is on the committee, but the idea came from Heber Blankenhorn, a staffer at the new National Labor Relations Board who needed evidence that the Wagner Act's promises were being ignored. It took the senator's gavel and a staffer's persistence to turn scattered complaints into a public record. That record helped give force to a principle the National Labor Relations Act still enforces. Employers may not spy on union activity, create the impression that they are spying, or photograph employees engaged in peaceful protected activity, according to the NLRB's own guidance.

The next federal limit targeted a single device. By the late 1980s, American employers were administering about two million polygraph tests a year, roughly 90 percent of them in the private sector. The Employee Polygraph Protection Act of 1988 barred most private employers from using lie detectors, eliminating about 85 percent of the tests then being given. Government employers, security firms, and drug manufacturers were exempted. Employers responded by switching to written honesty tests and expanded background checks, which the law did not cover.

American law in this area has grown one scandal and one technology at a time: the stopwatch (in federal workplaces only), the labor spy (only where union activity is involved), and the polygraph (one device). Outside those boundaries, an employer that owns the computer, the phone, the network, or the building has wide latitude to watch what happens on it. Until recently, only a handful of states, including Connecticut, Delaware, and New York, even required employers to tell workers they were being monitored.

The stopwatch moves into the laptop

Remote work turned employee monitoring into a mass market. In April 2020, global demand for monitoring software more than doubled. By 2023, 51 percent of American workers told the American Psychological Association that their employer used technology to monitor them, a share that was similar for office workers, manual laborers, and customer-service staff. Today's tools log keystrokes, capture screenshots, track location, and roll it all into "active time" or productivity scores, often in real time.

Employers have legitimate reasons. Banks and brokerages must keep records of client communications. Companies holding customer data need to detect leaks. Trucking and warehouse operators use sensors to prevent accidents. And some employees do game the system: in 2024, Wells Fargo fired more than a dozen people in its wealth management unit after reviewing allegations of simulated keyboard activity. The "mouse jiggler" may be the defining gadget of the era: an inexpensive device built to satisfy a metric that measured motion rather than work.

The evidence on what monitoring does to people is less flattering to the tools. Among workers in the APA survey, those who were monitored were twice as likely to say they felt uncomfortable at work (46 percent versus 23 percent) and more likely to feel micromanaged (51 versus 33 percent). Researchers Chase Thiel and colleagues found across two studies that monitored employees were more likely to break rules, including cheating on tasks and deliberately working slowly. Their explanation echoes Watertown: being watched shifted employees' sense of responsibility for their conduct onto the person doing the watching. The effect weakened when employees felt they were treated fairly, and simply explaining the scope and purpose of monitoring raised acceptance by about 70 percent.

Policy has swung with each administration. In 2022, the NLRB's general counsel issued a memo proposing that employers disclose the technologies they use to monitor workers and justify their necessity. In February 2025, the acting general counsel rescinded it along with dozens of other memos. Congress has not passed comprehensive rules on workplace AI or monitoring, so the states are again drawing their own lines, as they did with non-competes.

Connecticut, which has required notice of electronic monitoring since 1998, expanded its law on October 1, 2026. Employers there must now specify where on the premises monitoring occurs, and new hires must receive a plain-language statement of which prohibited activities may be monitored without notice. California went further. On September 30, 2026, it enacted a package that bans surveillance tools in restrooms (AB 1331), restricts AI tools that infer workers' emotional states or collect neural data (AB 1883), and, starting July 2027, bars employers from relying solely on an automated system to discipline or fire workers (SB 947). Under SB 947, a human must corroborate the decision and the worker must be told (KQED).

California's path shows the trade-offs. Governor Newsom vetoed a broader version in 2025, saying it imposed overly broad restrictions on how employers could use automated tools. The narrower 2026 law passed over the objections of the California Chamber of Commerce and other business groups, and it does not cover rideshare drivers, who are classified as independent contractors and have lived under algorithmic management longer than almost anyone. Readers of our September post on the fissured workplace will recognize the gap: protections attach to employees, while the most closely managed workers are often not employees at all.

What other countries are doing

Other countries start from a different premise: workers should have a say in how they are watched, or at least the monitoring should be proportionate to its purpose.

In Germany, the works council has a legal right of co-determination over the introduction and use of any technical system capable of monitoring employees' behavior or performance, under Section 87(1) No. 6 of the Works Constitution Act. Courts read the provision so broadly that German lawyers call it the works council's "Swiss army knife," and even a routine software update can require the council's agreement. Employers bear a real cost in slower rollouts and lengthy negotiations. In return, monitoring that reaches the floor has usually been negotiated with the people being monitored.

France shows what proportionality looks like in practice, including its limits. In January 2024, the French data protection authority (CNIL) fined Amazon's French warehouse unit €32 million for scanner-based indicators that flagged idle time to the second and penalized items scanned too quickly. The regulator did not object to monitoring as such; it objected to keeping every data point for a month when real-time data and weekly statistics would have done the job. In December 2025, France's highest administrative court cut the fine to €15 million and upheld much of Amazon's algorithmic management system. The line between managing work and monitoring workers is contested even in a country that takes it seriously.

The European Union has moved from data protection to AI-specific rules. Since February 2025, the EU AI Act has prohibited AI systems that infer employees' emotions in the workplace, with exceptions for medical and safety uses. Obligations for high-risk AI in hiring and worker management, including informing workers before such systems are used, were pushed back to December 2027. Separately, the EU's Platform Work Directive, which member states must adopt into national law by December 2, 2026, requires digital labor platforms to tell workers about automated monitoring and decision-making systems and bars them from processing certain kinds of personal data. California's new ban on emotion-inferring workplace tools brings one American state close to the EU position.

Ontario, which we discussed last month for its non-compete ban, chose transparency over restriction. Since 2022, employers with 25 or more employees must have a written electronic monitoring policy that states whether, how, and why they monitor. The law creates no new privacy right and places no limit on how employers use what they collect. It is easy to criticize as disclosure without teeth, but it has made every covered employer answer, in writing, the question many American employers have never asked themselves.

The approaches differ in cost and strength. Their shared principle is that monitoring needs a stated purpose, and that the people being monitored should know what it is.

AI: from watching the work to learning from it

Algorithmic management is Taylor's system with the stopwatch running all day. Software assigns tasks, sets the pace, scores the output, and in some workplaces flags people for discipline. The standard is no longer set by a young engineer standing at a molder's elbow; it is set by a model trained on aggregate data that no individual worker can inspect. California's new requirement that a human corroborate any AI-driven firing is, in effect, a demand that someone who can be questioned stand behind the stopwatch.

AI has also created a new thing to measure: whether people use AI. In April 2025, Shopify's CEO told employees that using AI was a baseline expectation, added AI usage questions to performance and peer reviews, and required teams to show that AI could not do the work before asking for more headcount. A few months later, a Microsoft division president told managers that AI use should be part of their assessment of each employee's performance, and the company was reported to be weighing formal usage metrics. The business logic is clear: companies that pay for AI tools want them used. But a count of prompts sent is to AI what keyboard activity was to remote work. Like a keystroke count, it rewards visible activity, and people quickly learn to supply it. Critics quoted by UNLEASH warned of exactly that: employees using tools to demonstrate compliance rather than to do better work.

The larger change concerns what happens to the data. Ford's investigators gathered information to decide whether a worker deserved the full wage. Today's systems gather information about the work itself, and what they learn can outlast the job. Nicole Moore, president of Rideshare Drivers United, told KQED that drivers' skills and experience have been turned into data that helps power autonomous vehicles, with no compensation and no say in how it is used. Readers of last month's post will recognize the question. Non-compete law asks what a worker may take when they leave; monitoring data, fed into AI, raises the question of what the employer keeps. A keystroke log used to be evidence about a worker. It can also become raw material for a system that can do part of that worker's job.

Most monitoring and AI adoption decisions are made by managers trying to solve real problems (security, quality, coordination across remote teams). The design choices still matter, and HR should not make them alone. Finance, legal, technology, operations, and HR each see a different part of the risk and return equation. Data can be collected about individuals or in aggregate, kept for a day or for years, used to coach people or to rank them, shown to workers or hidden from them. Each of those choices affects cost, risk, trust, and whether employees experience a tool as support or as suspicion. .

Worth: what is being watched, and what is it worth?

Monitoring is usually justified as measurement, and that is where it most often fails. Keystrokes, active minutes, badge swipes, and prompt counts are measures of activity. They tell a manager that something happened, not whether it created value. Taylor at least timed the work itself. Many of today's tools time proxies for work, and people respond to proxies by producing more of them. A human capital analytics team would never accept "hours logged" as a measure of training effectiveness; finance would not accept activity as evidence of return on investment. We should be equally skeptical of “active time” as a measure of contribution. 

In the work itself, what gets measured shapes how the job is done. Scanner-speed targets in a warehouse or handle-time targets in a call center change behavior, sometimes at the expense of quality or safety, so leaders should know what behavior a metric will reward before they add it. In the workforce, monitoring has historically fallen hardest on the people with the least power to object, from the Cold War civilian workers who were timed after the federal ban lapsed to today's app-based drivers outside employee protections. Who is monitored, and how intensively, is an equity question.

In the workplace, Ford sent investigators into workers' homes, and remote work brought the webcam and screen capture into the same kitchens. When the workplace is the home, the line between monitoring work and monitoring a life has to be drawn on purpose. As for worth, monitoring carries costs that rarely appear in the business case: software and administration, legal exposure across jurisdictions (as Amazon learned in France), false positives that require investigation, lower trust, and, according to the research above, more rule-breaking. Meanwhile, most organizations never measure whether monitoring improved the outcomes it was bought for, such as quality, safety, or customer results, productivity, or financial performance. That is an HR problem and a finance problem: if the organization cannot connect the metric to an outcome that matters, it cannot know whether the monitoring creates value. 

This is also a governance issue. Monitoring data is among the largest and most sensitive datasets a company holds about its people, and it is now also used to train AI. Boards and leadership teams should be able to answer: How material is human capital to our business model and enterprise value, and are we measuring the things that actually create that value? What do we monitor, and for what stated purpose? Which decisions (pay, promotion, discipline, termination) use monitoring data or AI-generated scores, and is a human accountable for each one? How long do we keep the data, who can see it, and can workers see their own? Is any of it being used to train systems that will automate parts of the same jobs, and if so, how is that contribution recognized? Have we measured whether monitoring improves outcomes, or only whether it produces data? And can we explain the return, the risk, and the trade-offs to investors, employees, and regulators? 

A century ago, IRC4HR's founders argued that workers are assets rather than costs. An asset you watch constantly without understanding how it creates value is a curious kind of asset. The organizations that get this right will measure outcomes rather than motion, connect workforce measures to business and financial results, explain what they monitor and why, and give the people being measured some visibility into the results.

Your Sphere of Influence

What you can act on	Practical solutions
Inventory	List every tool that collects data about employees: security software, productivity trackers, call recording, cameras, badge systems, telematics, and the AI assistants that log prompts. IT, security, operations, and individual managers often buy these separately, and HR may not know half of them exist.
Purpose test	For each tool, write down the specific business purpose and check whether a less intrusive option (aggregate data, shorter retention, sampling) would serve it. If no one can state the purpose, that is a finding.
Jurisdiction	Map monitoring practices against Connecticut's expanded notice law (in effect October 1, 2026), New York's notice-and-acknowledgment rule, California's new restrictions on restroom surveillance and emotion inference, Ontario's policy requirement, and the EU AI Act. A single global monitoring policy is now a compliance risk.
Transparency	Tell people what is monitored, why, who sees it, and how long it is kept, before it starts. In one study, explaining scope and purpose raised acceptance by about 70 percent. Let employees see their own data.
Human decisions	Where monitoring data or AI scores feed discipline, termination, pay, or promotion, require a named human reviewer who can explain the decision. California will require this for discipline and termination from July 2027; there is no reason to wait.
AI usage metrics	If AI adoption enters performance reviews, measure what AI helped produce (quality, cycle time, new capability), not how often it was used. Usage counts invite the same gaming as keystroke counts.
Measurement	Track whether monitoring improves the outcomes it was bought for (safety incidents, quality, data-loss events, customer results, productivity). Weigh that against costs: licenses, investigation time, legal exposure, engagement, and turnover and other workforce effects. Connect the measures to business and financial outcomes so leaders can assess return, not just activity.
Line of sight	Help managers and employees see how measured activity connects to results that matter. When people understand what is measured and why, and how it contributes to business outcomes, monitoring can support coaching rather than suspicion.
Governance	Give boards and senior leaders regular reporting on what is monitored, which decisions use monitoring data or AI scores, how the data is retained and protected, whether it is used to train AI systems, and what evidence shows it is worth the cost. Include the materiality of human capital to the business model and the human, legal, operational, and financial risks created by the monitoring approach.

Conclusion

For more than a century, the argument over workplace monitoring has followed a recognizable sequence. A new tool makes work more visible, employers use it to manage, workers resist, and the law eventually draws a narrow line around the most intrusive use. Congress banned the stopwatch in federal workplaces, exposed the labor spy, and outlawed the polygraph for most private employers, each time after a public controversy and each time around a single technology.

AI changes the scale. It can make nearly every kind of work visible at once, and it can turn what it observes into a system that performs part of the work. That adds a second question to the old one about how closely an employer may watch: what may an employer do with what it learns and how should leaders account for the value and risk created in the process?

Taylor warned the Watertown officers that time study imposed on unprepared workers would provoke resistance. They went ahead anyway and got a strike, a congressional hearing, and a ban that lasted about thirty years. Employers introducing AI monitoring today have the benefit of his warning, and of a century of evidence that people who understand what is measured, and why it is measured, and how it connects to outcomes are far more willing to accept it.

The lesson is not to measure less. It is to measure what matters, govern it well, and be able to show that the value created exceeds the human and financial cost. 

Next month

Why does your health insurance come from your employer? In 1933, a young surgeon named Sidney Garfield began offering prepaid medical care to workers building the Colorado River Aqueduct in the Mojave Desert, for a nickel a day. A decade later, wartime wage controls made health benefits one of the few ways employers could compete for scarce workers. We look at how an accident of history tied American health care to the job, what that has meant for workers who change jobs or work outside traditional employment, and what happens to the arrangement as AI reshapes who is employed.

---

This is the November 2026 installment of "What Have We Learned?", a monthly series for HR leaders on the history of the employer-worker compact.