The current understanding of automation is dominated by “routine-biased technological change” (RBTC). This theory predicts a strong automation dynamic in jobs with high routine-task share and a polarization of employment structures. While RBTC theory has many merits, this paper develops a systematic critique of the theory and a counter-proposal of a socioeconomically grounded company-level theory of the automation of work. It distinguishes between feasibility conditions of automation, technology choices, and social outcomes. With regard to feasibility conditions, the relevant factor is not routine-task intensity but the interaction between product architecture (product complexity) and process complexity. Which technology choices are made in this feasibility space is in turn influenced by companies’ profit strategies and power relations between management and labor. The social outcomes of automation depend on these technology choices, but also on managerial strategies pursued in the restructuring of organizational roles and skills. These managerial strategies are shaped by national institutional systems.