Principal-agent problems are fundamental to understanding many strategic interactions in practice. The economics literature has provided a rich theoretical foundation for principal-agent models, and the recent increase in attention from the computer science community has introduced fresh insights and new challenges. In addition, as we build upon these foundations laid by traditional approaches, we recognize that some may benefit from further refinement to capture the subtleties of agent decision-making, including bounded rationality and strategic deception, which can have a crucial impact on principal-agent relationships. By integrating perspectives from both economics and computer science, this dissertation aims to bridge the existing gaps and uncover a more comprehensive understanding of principal-agent problems with their real-world implications. First of all, we study a ubiquitous learning challenge in online principal-agent problems, where the principal learns the agent's private information from their revealed preferences in past interactions. This paradigm includes important special cases such as pricing and contract design, which have been widely studied in recent literature. Then we develop computational results on computing dynamic and robust principal policies to account for the subtleties of agent decision-making, where we move beyond the myopic agent behavior model to accommodate strategic and dynamic interactions. Finally, we apply the framework to a diverse range of real-world applications, such as market design and Stackelberg security games.