This is the paradox of productivity gains with AI. According to a study conducted by two researchers and published in the columns of the Harvard Business Review, the adoption of generative AI leads less to a reduction in employee workload than to an intensification of work, which can lead to degraded quality and professional burnout. Let’s deep dive.
Sociologists Aruna Ranganathan and Xingqi Maggie Ye from the University of California, Berkeley, spent eight months (April to December 2025) in an American technology company with approximately 200 employees to study how AI modifies work habits. They conducted on-site observations two days per week and conducted about forty interviews with professionals from engineering, product, design, and operations.
Their finding? AI tools have not reduced work. They have systematically intensified it!
We found that employees were working at a faster pace, assuming a wider range of tasks, and extending work over more hours of the day, often without being asked,” they write.
While adding this important clarification:
The company did not impose the use of AI (although it did offer corporate subscriptions to commercially available AI tools). Of their own initiative, workers were doing more because AI made it possible to do more, and in many cases, this was intrinsically gratifying.”
How to Explain This Gap Between a Laudable Initial Objective and the Opposite Reality?
The researchers identified three mechanisms:
1 – Task Expansion
By reducing barriers to learning and execution, AI enabled employees to perform tasks that would have previously justified external help (or a refusal!). In short, they assumed responsibilities formerly held by others, and thus reduced their operational dependency.
Product managers or designers began writing code, researchers took on engineering tasks… AI made tasks dedicated to specialists accessible. What started as just “I just want to test to see,” quickly transformed into sustainable expansion of the scope of activities.
With side effects: engineers found themselves correcting and guiding the “vibe coding” work generated by their colleagues’ AI, invisibly adding to their workload.
2 – Blurring the Boundaries Between Work and Breaks
The second effect is more subtle and insidious. Starting a task becomes so easy with AI… that natural breaks are gradually eroded! No need for reflection time facing a blank page. Employees began slipping small tasks in during dinner, meetings, or while waiting for a file to load. The famous “I’m just doing one last quick request” before leaving so the AI can “work” while the employee interrupts themselves.
These actions were rarely seen as additional work, but over time, they produced a work day with fewer natural breaks and continuous involvement. The conversational style of requests further softened the experience; writing to an AI feels more like a discussion, which facilitates work spilling into evenings or early mornings without deliberate intention,” explain the researchers.
3 – Increased Multitasking
Finally, as a corollary of the previous point, workers adopted a new rhythm where multiple tasks are performed simultaneously. For example, writing code manually while letting AI generate an alternative in parallel. A sort of background work.
This feeling of having a new work “partner” masked another reality: increasing cognitive load, constant shifts in attention, and a permanent feeling of juggling (to verify AI’s work after all).
A Vicious Circle
By accelerating certain tasks, AI has increased expectations in terms of speed… and thus made employees even more dependent on AI!
Several participants noted that although they felt more productive, they did not feel less busy and in some cases felt busier than before,” the researchers summarize.
Nature abhors a vacuum.
A Problem for Employees but a Net Gain for Companies? Not Really, According to the Authors. This superficial “victory” actually hides significant risks. This silent intensification creates cumulative cognitive fatigue, judgment errors, and ultimately a progressive degradation of work quality. Not to mention burnout.
How to Make These Productivity Gains Sustainable? By adopting “good AI practices,” propose the researchers. That is, by defining “a set of intentional norms and routines that frame how AI is used, when it is appropriate to stop, and how work should and should not expand in response to new capacity.”
In concrete terms, this can translate to:
- Intentional breaks (moments of reflection before major decisions)
- Sequencing (regulating work progression and not just its speed)
- Human anchoring (sanctifying times for listening and human connection when AI can lock you into more solitary and autonomous work)
In other words, without intentional action from organizations, AI will allow doing more… at the risk of becoming addicted to this race for ever more.
“The question facing organizations is not whether AI will change work, but whether they will actively lead this change. Or whether they will let AI change them silently,” conclude the researchers.
training.isarta.com 
