A number of factors, including design variation and the combination of uncertain costs and cost-plus contracting, diminished opportunities and incentives to improve power plant design and construction over the last several decades. This paper incorporates these factors into a model of learning that relies on a principal-agent framework. I find that (1) because of design variation learning was reduced when an agent contracted with a series of different principals, (2) agent learning declined when cost uncertainty increased during the late 1970s and 1980s, and (3) at the same time the locus of learning may have shifted from agents to principals.