In collaborative settings, difficulties in sustaining a consistent pace and engagement often lead to task drift, reducing preparedness and overall effectiveness between meetings. To address this challenge, we conducted a formative study and developed ReflecToMeet, an AI assisted system that integrates theory driven reflective prompts with mechanisms for sharing teammates reflections. Informed by ten formative interviews, the system was evaluated in a mixed method study across three conditions: deeper reflection, regular reflection, and a control condition with unstructured reflection. Participants in the control condition demonstrated less deliberate thought and weaker collaboration, which led to stress and misalignment during team meetings. In contrast, structured reflection supported greater organization and steadier progress. The deeper reflection condition further facilitated confidence, teamwork, and idea generation, although it imposed a higher cognitive load. We conclude by discussing design implications for AI agents that facilitate reflection to enhance collaboration and broader considerations for AI assisted systems aimed at sustaining collaborative goals.

翻译：在协作环境中，难以维持一致的节奏和参与度常常导致任务偏离，降低会议间的准备度和整体效率。为应对这一挑战，我们进行了一项形成性研究并开发了ReflecToMeet，这是一个AI辅助系统，它将理论驱动的反思提示与分享队友反思的机制相结合。基于十次形成性访谈的启示，该系统通过一项混合方法研究在三种条件下进行了评估：深度反思、常规反思以及作为对照的非结构化反思。对照条件下的参与者表现出较少的深思熟虑和较弱的协作，导致团队会议期间出现压力和目标不一致。相比之下，结构化反思支持了更好的组织性和更稳定的进展。深度反思条件进一步促进了信心、团队合作和创意生成，尽管它带来了更高的认知负荷。最后，我们讨论了促进反思以增强协作的AI智能体的设计启示，以及旨在维持协作目标的AI辅助系统的更广泛考量。