Researchers Tianyi Fu and Mohan Sridharan have proposed a new architecture for assistive AI agents that addresses ambiguity in object references. Their approach, called hierarchical compositionality, represents objects as combinations of primitive attributes and concepts identified from human-validated semantic feature norms. The agent then uses these representations to reason about domain dynamics and user preferences, requesting clarification when necessary. Experiments show that this method outperforms state-of-the-art data-driven baselines in adapting to specific user profiles. --- Why it matters: This work matters because it offers a more principled approach to developing assistive AI agents that can handle ambiguity in object references, which is an important challenge in human-AI interaction. The proposed architecture has the potential to improve the reliability and effectiveness of such agents. Source: https://arxiv.org/abs/2608.10330