We examine a learning-by-doing methodology for iteration of early-stage business ideas known as the “lean startup.” The purpose of this article is to lay out and test the key assumptions of the method, examining one particularly relevant boundary condition: the composition of the startup team. Using unique and detailed longitudinal data on 152 NSF-supported lean-startup (I-Corps) teams, we find that the key components of the method—hypothesis formulation, probing, and business idea convergence—link up as expected. We also find that team composition is an important boundary condition: business-educated (MBA) members resist the use of the method, but appreciate its value ex post. Formal training in learning-by-thinking methods thus appears to limit the spread of learning-by-doing methods. In this way, business theory constrains business practice.