Proactive Execution in LLM-based Agent Systems: A Survey of Triggers, Mechanisms, and Challenges

Most LLM-based agent systems operate reactively, taking action only after receiving explicit user instructions. Yet many real-world settings demand that agents anticipate user needs, detect environmental changes, and correct their own reasoning before problems surface---capabilities we collectively term proactive execution. These capabilities are particularly relevant to consumer technologies and electronics, where proactive LLM-based agents can support smartphones, smart homes, wearables, personal robots, and IoT devices. However, related research remains scattered across proactive dialogue, embodied perception, feedback-driven adaptation, and self-reflection, with no unified framework connecting them. This survey bridges that gap. We propose a trigger-source taxonomy that organizes proactive execution into three paradigms: (i) user behavior-driven, encompassing need inference from interaction history, task-driven clarification, and assistance under intent shifts; (ii) external event-driven, covering environmental, behavioral, and informational event perception as well as feedback-driven closed-loop adaptation; and (iii) internal uncertainty-driven, addressing confidence-aware verification and self-reflection-based correction. For each paradigm, we analyze representative methods, compare their mechanisms and limitations, and identify the conditions under which they are most effective. We further extend the discussion to the Internet of Agents (IoA) setting, examining how proactive coordination scales across heterogeneous, networked agent infrastructures. Finally, we outline open challenges, including the trade-off between proactive intervention and user disruption, security and privacy risks, and the lack of evaluation standards for proactive behavior, and propose concrete directions for future research.