Does AI Adoption Increase Work Intensity and Monitoring?

Does AI itself make work more intense?

Work intensity means the demands placed on a worker: pace, volume, time pressure, breaks, interruptions, or extra checking and coordination. A writing assistant that shortens a first draft changes a task. Whether the whole job becomes more demanding depends on what happens next. The saved time might allow careful review, reduce overtime, support more complex work, or be used to increase the number of cases due each day.

The OECD’s 2023 Employment Outlook reports that three quarters of surveyed AI users in finance and manufacturing said AI increased the pace at which they performed tasks. That is an important signal about workers’ experience, but it is not a measurement of total workload or a causal estimate for all workplaces. The underlying OECD survey covered 5,334 workers and 2,053 firms in those two sectors across seven countries. People were asked how AI changed their work; the answers should not be treated as a universal result for today’s generative tools or every occupation.

A separate OECD analysis of Japan combines international comparisons with a 2024 Japan Institute for Labour Policy and Training (JILPT) worker survey. Among 1,854 Japanese AI-user respondents in the cited results, workers reported improvements in performance and aspects of job quality more often than deterioration, while other parts of the analysis describe faster task pace and concerns about data collection. These are workers’ retrospective perceptions, not a causal experiment; the Japan findings should not be generalized to all workers, sectors, or current generative-AI deployments. Benefits and pressure can coexist: someone may finish a task faster and find it more satisfying while also facing shorter deadlines or more tasks. Perceived productivity, task speed, total output, work quality, and pressure are different outcomes.

For your own work, compare the task before and after introduction: its duration, the amount expected, the review standard, response time, and the breaks available. A faster task is evidence of changed capability or workflow. A sustained rise in volume or pace is evidence of a changed demand. The second does not follow automatically from the first.

Sources: Artificial intelligence, job quality and inclusiveness: OECD Employment Outlook 2023; Reaping the benefits of AI for performance at work and job quality: Artificial Intelligence and the Labour Market in Japan

How can efficiency gains become pressure?

Efficiency can become pressure when an organization turns quicker completion into more work, tighter turnaround, fewer staff for the same queue, or an expectation of constant availability. This is a management pathway, not an inevitable property of the tool. A system may expand capacity while a team decides how to distribute the gain: keep output steady and improve quality, take on a wider mix of work, reduce backlogs, or raise throughput targets.

OECD case studies describe workers moving toward more complex tasks and warn that automating easier tasks can remove the mental breaks those tasks provided. This evidence comes from semi-structured interviews with workers, managers, and others at 90 firms across eight countries, mainly in finance and manufacturing. Researchers recruited firms using their own networks and contacts, so these selected cases illustrate possible mechanisms; they are not a representative estimate of how often AI causes stress. Some cases also involved technologies other than AI. They show how redesign can change demands, not that AI alone caused pressure across workplaces.

A demanding new task can also be both more engaging and more stressful. A support worker whose routine replies are drafted automatically may spend more time on difficult cases. That can be a useful shift if the volume is manageable and the worker has discretion, training, and time to resolve exceptions. It can be a strain if difficult cases arrive at the old rate while the freed capacity is filled with additional cases. The task mix, staffing, and pace together decide which pattern is closer to reality.

So do not infer worse work from an efficiency claim, and do not infer better work from a productivity figure. Look for observable changes: cases or orders handled, queue length, deadline windows, interruptions, review and correction time, staffing, and missed breaks. Ask whether a target changed and whether the quality measure changed with it. A time saving only describes capacity; it does not reveal who receives that capacity.

Sources: Artificial intelligence, job quality and inclusiveness: OECD Employment Outlook 2023

Is monitoring part of AI adoption, or a separate choice?

They are distinct and can occur separately. A worker may use AI to summarize documents without any new tracking. An employer can monitor working time or allocate shifts through ordinary digital systems without generative AI. Algorithmic management is a broader term for computer-programmed procedures that coordinate or direct work; it may include AI, but the two terms are not interchangeable.

The JRC’s AIM-WORK survey, conducted in 2024–25 with 70,316 workers across all 27 EU Member States, measured AI use, digital monitoring, and algorithmic management as distinct features. Its summary reports work-related AI use by 30% of workers and working-time monitoring for 37%; automatic allocation of working time was reported by 24%. These are population survey estimates, not evidence that AI users are necessarily the people monitored or that AI caused monitoring. The survey is a broad snapshot of digital work, not a causal test of a particular AI tool.

The JRC report on the same survey distinguishes patterns of ‘platformisation’: some types show no significant implications for working conditions, while full platformisation—combining all identified forms of digital monitoring and algorithmic management—is associated with generally worse conditions. The accessible summary does not support a more specific split into informational and physical effects, and the association does not establish causation. Separately, an OECD survey of more than 6,000 firms in France, Germany, Italy, Japan, Spain, and the United States asked employers about algorithmic-management practices, perceived impacts, and governance. This describes employer-reported practices and views, not workers’ experiences or causal effects on job quality. Recording an event, evaluating performance, and assigning the next task are different forms of control.

If a new metric appears, ask what data it collects, whether a person reviews it, what decision it informs, how errors or context can be corrected, and whether it merely records work or sets its pace. That is more useful than calling every digital metric AI surveillance. The distinction also points to different remedies: task design and output expectations for workload; transparency, access, correction, and limits on data use for monitoring.

Sources: Algorithmic management in the workplace: New evidence from an OECD employer survey; Algorithmic management and digital monitoring of work; Impact of digitalisation: 30% of EU workers use AI; Digital Monitoring, Algorithmic Management and the Platformisation of Work in Europe

What would show which pattern is happening in your job?

Track task change and management control on separate lines. For a typical week, note the kinds of work you do, approximate case or project volume, turnaround expectations, rework, interruptions, and whether breaks are possible. Separately note what information is collected about your work, who can see it, and whether a tool ranks performance, allocates tasks, or determines pace. You do not need a perfect baseline; a simple before-and-after record can clarify what question to raise.

Treat time saved cautiously. It may be an estimate, and a shorter first pass may create new verification or coordination work. The evidence above measures reported task pace and digital controls separately; use that distinction when you compare your own before-and-after notes. Your record cannot prove what caused a change, but it can identify whether the immediate question is about volume and recovery time, data collection, or a system directing work.

A proportionate next step is a focused conversation with your manager: ‘Which output or response-time expectation changed when we introduced this tool, how are quality and exceptions measured, and what data about my work is collected or used to make decisions?’ If a target changed, suggest reviewing workload, quality, exception handling, and breaks together. If data are collected, ask who can access them and how errors or context can be corrected. When the answer remains unclear, keep dated examples and use the normal workplace process for workload or privacy concerns; relevant rights and procedures depend on where you work.

If you are also mapping which duties face task-level change, the free [task checker](/ai-job-risk-checker) can help; it measures task-level change pressure, not surveillance or displacement.

Sources: Artificial intelligence, job quality and inclusiveness: OECD Employment Outlook 2023; Algorithmic management and digital monitoring of work; Impact of digitalisation: 30% of EU workers use AI; Digital Monitoring, Algorithmic Management and the Platformisation of Work in Europe

Questions readers ask