Self-adaptability has been proposed as an effective approach to automate the complexity associated with the management of modern-day software systems. While over the past decade we have witnessed significant progress in the manner in which such systems are designed, constructed, and deployed, there is still a lack of consensus among the engineers on some of the fundamental underlying concepts. In this paper, we attempt to alleviate this issue by exploring the crucial role of computational reflection in the context of self-adaptive software systems. We show that computational reflection forms the foundation of a self-adaptive system, and an understanding of its properties is a prerequisite to intelligent and predictable construction of such systems. Examining several systems in light of computational reflection has helped us to identify a number of key challenges, which we report on and propose as avenues of future research.