Picture a consultant who asks an AI assistant to prepare a report just before leaving for the day. At home, they keep wondering whether the figures are right and whether an important qualification has quietly dropped out. In this hypothetical case the task has been delegated, but the responsibility for it has not been settled.
During International Stress Awareness Week, 2-6 November, this kind of unresolved responsibility deserves as much attention as talk of efficiency. A system can carry on processing information after its user has closed the laptop, and the person can stay mentally tied to whatever it produces.
Concerns about keeping people genuinely engaged when AI does the work usually focus on the quality of decisions. The same gap can also follow people home, and it is worth looking at what that does to recovery.
Finishing a task is not the same as recovering from work
Recovery is about how we feel and function after effort. Sabine Sonnentag and Charlotte Fritz describe psychological detachment as mentally disengaging from work during time off. Being away from the office says little about whether work is still occupying someone’s attention.
A meta-analysis by Andrew Bennett and colleagues looked at recovery experiences across 54 independent samples and linked recovery with how employees experience demands, resources, and well-being. Those findings help explain recovery, but they cannot prescribe one evening routine that suits everyone.
Thinking about work is not automatically harmful either. Someone may enjoy turning over a difficult problem or feel pleased about a finished project. The concern here is persistent, unwanted engagement that makes rest difficult, especially when the issue cannot be resolved until work starts again.
What changes when a system keeps working
A 2026 study by Zhang, Li and Chen on the vigilance cost of AI delegation surveyed 642 hybrid or mostly remote employees at two points in time. The authors examined the ongoing watchfulness people feel around delegated work, including worries about output quality and possible errors.
Their results linked this vigilance with work-related rumination and with difficulty disengaging from work. That association does not show that AI causes poor recovery, since the study relies on self-reports and its proposed mechanism still needs independent replication.
The distinction matters because a helpful tool can still sit inside an unclear working arrangement. One person might save effort producing a document while remaining unsure who checks it. Another might gain the freedom to finish early because responsibility and the next review point are explicit, which is part of why digital tools can create more work rather than less.
Our consultant cannot settle the practical question of who will check the report before anyone acts on it. Until that responsibility is assigned, telling them to switch off leaves the uncertainty exactly where it was.
Availability can turn into an expectation
This problem predates generative AI. A diary study by Daantje Derks and colleagues examined how work-related smartphone use relates to detachment and exhaustion, and it also considered workplace norms about keeping work and private life apart. That shifts attention to the context in which a device is used.
A message sent in the evening can simply inform, or it can create an expectation of action. Employees may read the same message differently, particularly when quick replies have been rewarded in the past. Stating the expected response time gives people something more useful than a general promise of flexibility.
The same question applies to automated activity. Who receives an alert, and who is expected to respond? If nobody has agreed the answer, each person may keep monitoring because they assume the responsibility is still theirs.
Make the next responsibility clear
I would start with one delegated task. Identify who will check the result and when that check should happen. Where genuine operational cover is needed, agree who provides it and how responsibility passes from one person to the next.
An end-of-day note can then record what has been completed and the next action required. Its purpose is to preserve what tomorrow needs, not to become another prompt to keep checking the same task all evening.
Managers can help by separating an ordinary update from a request that needs immediate action. They can also look at whether supposedly optional checking has become an unofficial performance expectation, because otherwise a person who respects the stated boundary may feel exposed when colleagues stay available.
These changes are about how work is organised. A review by Sonnentag, Venz and Casper shows that recovery depends on both individual activities and working conditions, so an exercise that helps one person may not be enough when demands stay excessive.
When work-related worry keeps disrupting sleep or everyday life, professional support can help someone understand what is keeping it going. A good assessment will consider their circumstances, including the working arrangements they cannot change alone.
For the next task you delegate, agree the next checking point before you finish for the day. That gives the technology a defined role and gives you a clear account of what remains your responsibility.
