AI assistants: saving time without fragmenting work
In SHR — Swiss Human Resources’ programme “Create your AI assistants and agents without coding”, design concerns not only what the tool produces, but also how it enters the workflow. The management question is precise: should people review each result as soon as it becomes available, or group their reviews to preserve continuity of work?
What the literature establishes
Sophie Leroy (2009, Organizational Behavior and Human Decision Processes) demonstrates experimentally that, after a task switch, some attention can remain attached to the previous task, particularly when it is unfinished. This “attention residue” can impair performance on the next task. For an AI assistant, the implication is a design hypothesis: an immediately available result is not necessarily one that should be reviewed immediately.
The cost of switching rules
Joshua S. Rubinstein, David E. Meyer and Jeffrey E. Evans (2001, Journal of Experimental Psychology: Human Perception and Performance) identify time costs when switching between tasks in controlled experiments. Their findings support an executive control mechanism involving goal shifting and activation of the relevant rules. Moving from a case file to an assistant’s response could therefore involve more than reading time alone, although this study does not directly measure that use.
The premature conclusion
One might conclude that notifications should be disabled and all results reviewed at fixed times. Yet these studies establish neither an optimal review frequency nor the general superiority of batching for AI assistants. An alert that determines whether work can continue may justify an interruption; a non-urgent draft can probably wait. (our executive and employee training programmes)
What the evidence cannot settle
These studies concern neither today’s generative agents nor their no-code interfaces. They cannot quantify an assistant’s net benefit within a team, where deadlines, dependencies between colleagues and deliverable quality also matter. Measuring generation time alone leaves out precisely the time spent reviewing, resuming work and making corrections.
A practical check in Genève
In Genève, an exercise in the SHR programme can compare two review arrangements using fictitious or authorised case files: reviewing each result as it becomes available, then reviewing at agreed times, with comparable tasks and the order reversed between participants. In private banking, an international organisation, luxury watchmaking or commodity trading, the test could respectively involve a draft client note, a meeting summary, an after-sales service response or a document overview, excluding urgent alerts. For each case, record the number of switches to the assistant, cumulative human working time through to approval and the corrections required under a common review rubric. Grouped reviewing should be retained only if observations show time savings without reduced quality or missed agreed deadlines: a local decision criterion, not a universal promise. To go further: explore the Building AI Assistants and Agents Without Coding training in Geneva, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Geneva
- building ai assistants and agents without coding
- Geneva
- research
- training geneva