A novel autonomous mobile GUI agent, UI-Voyager, has been proposed to address the limitations of existing methods in learning from failed trajectories and handling sparse rewards in long-horizon GUI tasks. This two-stage self-evolving agent is designed to improve the efficiency of learning from failed experiences, leveraging the advancements in Multimodal Large Language Models (MLLMs). The agent's capability to learn from failures and adapt to complex GUI tasks has significant implications for various applications, including those that require human-computer interaction. The development of UI-Voyager demonstrates the potential of autonomous agents in navigating complex graphical user interfaces, which can lead to increased automation and efficiency in various sectors1. This breakthrough matters to practitioners and informed readers because it highlights the potential of AI to transform the way humans interact with technology, raising important questions about the security, policy, and workforce implications of such advancements.