Emerging Augmented Reality (AR) technologies offer immersive, spatially anchored interfaces that can transform remote collaboration beyond the limitations of traditional 2D video conferencing. In this work, we introduce a task-agnostic AR system that dynamically adapts the placement of virtual elements and visual guidance cues based on real-time user interaction context and avatar behavior. Unlike prior approaches that statically anchor content to fixed locations or rely on manual placement, our framework continuously evaluates a sliding window of user focus and applies a multi-objective optimization, balancing proximity, visual attention, and collaboration context, to reposition interface elements. We evaluated our system on two representative tasks (problem solving and knowledge-work) with 10 participants, comparing four placement strategies (random, proximity-based, individual action-based, group action-based). Quantitative results show that our action grouping method reduces task completion time by up to 20 % and gives higher subjective ratings of focus and usability. Qualitative feedback further underscores the importance of low-frequency repositioning and context-aware grouping to minimize distraction. These findings demonstrate that adaptive AR interfaces can significantly enhance collaborative efficiency and user satisfaction, suggesting a new direction for context-driven UI in mixed reality environments.