The idea that the division of labor increases productivity is central to economic analyses of countries, industries, social structures, and occupations within organizations. We study the division of labor within an online primary healthcare organization, where an algorithm assigns patient cases between two clinician occupations: nurses and doctors. We compare a knowledge hierarchy, in which less specialized nurses resolve some cases themselves and escalate others to doctors, with a direct-to-expert approach where cases are assigned directly to doctors. We use approximately 500,000 cases and an identification strategy that leverages temporary congestion which increases the odds of a nurse-initiated knowledge hierarchy assignment. We find that nurses resolve 70% of cases at the margin and send the remainder to doctors. Although the knowledge hierarchy slightly reduces the rates of meaningful diagnosis and prescription, it does not adversely impact patient satisfaction, acute care utilization, labor earnings, or mortality. The knowledge hierarchy lowers total costs to the healthcare system without compromising quality at the primary care organization or downstream. Finally, we explore how the knowledge hierarchy’s comparative advantage varies across case types and assess the extent to which task allocation aligns with comparative advantage.