I study workers’ preference for automation in a setting where automation complements workers’ skills but is imperfect. While such technology increases workers’ productivity and wages by enabling work on a new task, it subjects workers to a potentially costly multitasking environment. I identify and test conditions that determine a worker’s preference for automation in this setting. My theoretical model borrows elements from a task-based framework and predicts that, under certain conditions, a worker’s preference will follow a threshold rule. The threshold is given by a parameter that determines how well a work environment is suited for automation. In a real-effort experiment, I exogenously vary this parameter and observe how this variation affects subjects’ choices of technology. I find that the treatment effect is in line with theoretical predictions: subjects choose automation more frequently in environments that are more suited for automation. Using machine learning techniques, I explore which subjects’ characteristics predict choices of technology and find that subjects’ task performance tends to be an important predictor. I discuss the implications of my results for workers’ welfare, technology adoption, and inequality.