Abstract: With the evolution of satellites from information acquisition terminals toward on-orbit intelligence, satellite software has become the core foundation for achieving autonomous perception, intelligent decision-making, and coordinated control. Traditional satellites mainly undertake data acquisition and transmission tasks, whereas new-generation low-orbit satellites are gradually being equipped with high-performance processors, FPGA modules, and accelerators, enabling them to execute complex application tasks in orbit, such as data analysis and processing, low-latency emergency communications, and autonomous mission scheduling and planning. This in-orbit computing paradigm has significantly improved the real-time processing capability and mission autonomy of satellite systems, and has driven a transformation from ground-dependent architectures to in-orbit intelligent architectures. However, due to constraints, such as limited in-orbit resources, communication delays, environmental uncertainty, and security requirements, the development, evolution, and maintenance of satellite application software face a series of new challenges, including intelligent optimization under resource-constrained conditions, the feasibility and reliability assurance of in-orbit updates, and in-orbit recovery from software failures. This article explores the key challenges and opportunities involved in realizing intelligent application software in satellite computing scenarios, covering multiple stages, including software design, execution, evolution, and maintenance. Finally, taking satellite application software update as an example, we analyze its in-orbit characteristics and technical difficulties, propose the authors’ solution for in-orbit application software updates, and demonstrate a significant improvement in effectiveness.