AI doesn't cut down people's workload, but ends up making them busier.

At present, many enterprises are struggling with how to get more employees to use AI. After all, the prospect of AI application is extremely attractive: it can reduce the workload by drafting routine documents, summarizing information, debugging codes and other methods, so as to free up more time for employees to handle high-value work.

But if leaders successfully get more employees to use AI, are they prepared for everything that comes next? When leaders only focus on the potential productivity improvement, the complex reality brought by AI may catch them off guard. By the time they realize the cost behind these achievements, it may be too late.

In an ongoing research, we found that AI tools do not reduce the workload, but instead increase the work intensity. We conducted an 8-month study on an American technology company with about 200 employees to examine how generative AI changes employees' work habits. The results show that compared with employees who do not use AI, employees who use AI work at a faster pace, take on a wider range of tasks, and work longer hours, and in many cases, the company does not require them to do so. Importantly, this company does not force employees to use AI (although the company does provide enterprise subscriptions for commercial AI tools for employees). Employees voluntarily take on more tasks because AI makes "doing more" feasible and easy, and in most cases, this initiative can also bring a sense of intrinsic satisfaction.

This may sound like a dream situation for leaders, but the changes brought by employees' active use of AI may not be sustainable, and in the long run, it will trigger a series of problems. When the enthusiasm for trying new tools fades, employees will find that their workload has increased unconsciously, and they will feel overwhelmed when facing all kinds of tasks that suddenly pile up in front of them. The accumulating workload may further lead to cognitive fatigue, occupational burnout, and reduced decision-making efficiency. The initial surge in productivity may eventually evolve into other problems such as declining work quality and employee turnover.

This puts leaders in a dilemma. What should they do? Asking employees to adjust on their own is not a good strategy. On the contrary, companies need to formulate a set of norms and standards around the use of AI, that is, "AI practice". Leaders can take the following ways to help employees use AI successfully.

How Generative AI Increases Work Intensity

In this study, we examined how generative AI tools change the work habits of employees in technology companies. We conducted on-site inspections two days a week, tracked internal corporate communication channels, and carried out more than 40 in-depth interviews with employees from engineering, product, design, research and operation departments. The study found that the increase in work intensity is mainly reflected in three aspects.

1. Wider range of tasks. Since AI can fill knowledge gaps, employees begin to take on responsibilities that originally belong to others. Product managers start to write codes, researchers take over engineering tasks, and many people in the organization try jobs that they would outsource, postpone or avoid completely in the past.

Generative AI makes these tasks easier than ever before. AI tools empower many people and enhance their cognition, reduce employees' dependence on others, and provide instant feedback and correction suggestions in the process. Employees describe this experience as "just casually trying with AI", but these attempts actually expand their scope of work. In fact, employees take on more work that would otherwise require additional temporary helpers or more staff to complete.

The expansion of the scope of responsibilities has brought a series of ripple effects. For example, engineers need to spend more time reviewing, correcting and guiding the work of colleagues, because these works are usually generated or assisted by AI. These demands go beyond the scope of formal code review. More and more engineers find themselves needing to guide colleagues who are engaged in "vibe coding" and help them finish their unfinished work. This kind of supervision often occurs in informal scenarios, such as follow-up comments on Slack or quick consultations next to the workstation, which further increases the workload of engineers.

2. The boundary between work and non-work is blurred. Because AI makes it very easy to start a task and eliminates the pressure of not knowing how to start, employees squeeze a small amount of work into their break time. Many people send instructions to AI during lunch or meetings. Some people send one last quick instruction before leaving their workstation, so that AI can continue to work after they leave the computer.

These behaviors rarely make people feel that they have done more work, but over time, this reduces the break time during the workday and keeps people more continuously engaged in work. Entering instructions to AI in a conversational way does not feel like working. In other words, typing a line of text into the AI system feels more like chatting than performing a formal task, which makes it easy for work to unknowingly extend into the night or early morning.

Some employees often realize afterwards that as they get used to entering prompts to AI during breaks, the break time no longer restores their energy as before. As a result, work no longer has clear boundaries and becomes ubiquitous, making people want to push a little further at any time. In fact, the boundary between work and non-work has not disappeared, but it has become easier to cross.

3. More multi-tasking. AI has brought a new work rhythm, and employees now need to manage multiple parallel task threads at the same time: manually writing codes while letting AI generate another version of the code; running multiple agents at the same time; or restarting tasks that have been put on hold for a long time, because AI can handle them in the background. The reason why they can do this is partly because they feel that they have a "partner" who can help them advance the work at hand.

Although this feeling of "having a partner to help" creates work motivation, the reality is that employees have to continuously shift their attention, frequently check the output of AI, and face more and more unfinished tasks. This creates a cognitive load, leaving people in a state of being overwhelmed, even though the work itself seems very efficient.

Over time, this work rhythm raises people's expectations for speed, which does not come from clear requirements, but from the phenomena that gradually emerge in daily work and are regarded as normal. Many employees pointed out that compared with the state before using AI, they now need to handle more tasks at the same time and feel greater pressure, even though the original intention of saving time in an automated way was supposed to reduce this pressure.

How "AI Practice" Benefits Organizations

All of this forms a self-reinforcing cycle. AI accelerates the processing of certain tasks, thereby raising people's expectations for speed; the improvement of speed makes employees more dependent on AI; and the deepening of dependence on AI broadens the scope of work that employees try to complete, which further increases the workload and intensity. Many research participants pointed out that although they feel that their work efficiency has improved, they do not feel more relaxed, and in some cases even feel busier. As one engineer summed it up: "You thought that with AI, you could be more efficient and save some time, so that you could do less work. But in reality, the workload has not decreased. You have as much work as before, or even more."

Enterprises may regard this voluntary behavior of employees to take on more work as a surefire good thing. After all, since employees take the initiative to propose it, what is wrong with that? Isn't this exactly the productivity boom we have been expecting?

But our research shows that when work continues to expand and accelerate without norms, it will bring some risks: the seemingly higher productivity in the short term may actually hide the quietly growing workload and the accumulating cognitive pressure of employees when they deal with multiple AI workflows at the same time. Since this extra input is often voluntary and is often regarded as a "pleasant attempt", it is easy for leaders to ignore how heavy the extra burden that employees are bearing is. Eventually, overwork will damage judgment, increase the possibility of errors, and make it more difficult for organizations to distinguish real productivity gains from unsustainable high-intensity work. For employees, the cumulative effect is manifested as fatigue, burnout, and an increasingly difficult feeling to step away from work, especially when the organization's expectations for speed and responsiveness continue to rise.

For both enterprises and employees, instead of passively facing the reality that AI tools reshape the workplace, it is better to actively adopt a set of "AI practice", that is, a series of specially formulated norms and conventions to clarify how AI is used, when it should stop being used, and how the scope of work should be expanded or avoided based on this new capability. In the absence of such "AI practice", the natural trend of AI-assisted work will not weaken, but will further intensify, which will have an impact on occupational burnout, decision-making quality and long-term sustainability.

When building their own "AI practice", enterprises should consider taking the following measures.

Conscious pause. As the pace of tasks continues to accelerate and work boundaries become increasingly blurred, enterprises can adjust the work rhythm by providing employees with short and structured time. During this uninterrupted period of time, employees can evaluate the consistency of work direction, re-examine existing assumptions, and fully digest information before continuing to advance their work.

These pauses will not slow down the overall work progress, they are only designed to avoid the quiet increase of workload in the unrestrained acceleration process. For example, before making a major decision, the organization can ask the team to take time to put forward a counterargument and clearly point out the connection between the decision and the organizational goals, so as to widen the scope of attention slightly and prevent the decision from going off track. Incorporating such pause links into the daily work process helps organizations make more informed decisions, establish healthier boundaries, and form a more sustainable productivity model in an AI-augmented environment.

Prioritize tasks. Considering that AI can run continuously in the background, enterprises should not only pursue speed, but also formulate relevant norms to consciously grasp the work rhythm. These norms include batching non-urgent notifications, waiting for work gaps to synchronize the latest information uniformly, and protecting employees' focus time from interference.

Organizations that prioritize tasks will encourage employees to advance their work in coherent and orderly stages, instead of requiring employees to respond immediately to every AI-generated result. In this way, employees' working hours are less fragmented, and the time and energy consumed by switching between different tasks are reduced. At the same time, the entire team can also maintain high work efficiency. By managing the sequence and timing of work, rather than requiring employees to respond continuously, organizations can better protect employees' attention resources, reduce cognitive load, and make more prudent decisions in an AI-forward work environment.

Maintain interpersonal connections. With the support of AI, more and more work can be completed independently by individuals. It will be of great benefit for enterprises to reserve time and space for employees to connect with each other. Whether it is through quick information synchronization, common review moments, or structured discussions, these short opportunities to connect with others can break the state where employees continuously use AI tools alone, and help them regain a comprehensive perspective.

Social interaction can also stimulate creativity. AI provides an integrated single perspective, while real creative insights come from diverse human perspectives. By reserving time and space for listening and dialogue, organizations can bring work back to the connection between people, and alleviate the energy consumption of individuals caused by the fast-paced AI-driven work.

The application prospect of generative AI lies not only in what it can bring to work, but also in how enterprises prudently integrate it into the daily rhythm. Our research found that without conscious design, AI will make people do more, but it is harder to stop. "AI practice" plays a balancing role: even if the work pace is accelerating, enterprises can reserve time for employees to restore energy and review. The real problem facing enterprises is not whether AI can change work, but whether they want to actively shape this change, or let AI change themselves quietly.

Aruna Ranganathan, Xingqi Maggie Ye | Article

Aruna Ranganathan is an associate professor at the Haas School of Business, University of California, Berkeley. Xingqi Maggie Ye is a doctoral student in organizational management at the Haas School of Business, University of California, Berkeley.

This article is from the WeChat official account "Harvard Business Review" (ID: hbrchinese), author: HBR-China, editor: ZHOU Jingyi, published with authorization from 36Kr.