As organizations and individual workers increasingly adopt generative AI (GenAI) to improve productivity, there is limited understanding of how different modes of human-AI interactions affect worker experience.

In this study, we examine the ordering effect of human-AI collaboration on worker experience through a series of pre-registered laboratory and online experiments involving common professional writing tasks. We study three collaboration orders: AI-first when humans prompt AI to draft the work and then improve it, human-first when humans draft the work and ask AI to improve it, and no-AI. Our results reveal an important trade-off between worker productivity and worker experience: while workers completed the writing draft more quickly in the AI-first condition than in the human-first condition, they reported significantly lower autonomy and efficacy. This negative ordering effect affected primarily female workers, not male workers.

Furthermore, being randomly assigned to a collaboration mode increased workers’ likelihood of choosing the same mode for similar tasks in the future, especially for the human-first collaboration mode. In addition, writing products generated with the use of GenAI were longer, more complex, and required higher grade levels to comprehend. Together, our findings highlight the potential hidden risks of integrating GenAI into workflow and the imperative of designing human-AI collaborations to balance work productivity with human experiences.