🖐️ What if an AI agent's hands didn't just point, but shaped, mimicked, and emphasized what it was saying? Here's research on agent conversations in XR. Title: AgentHands: Generating interactive hand gestures for spatially grounded agent conversations in XR URL: ❓ Why isn't a voice-only agent enough? 💡 Traditional 2D bounding-box overlays are built for flat screens and don't translate to immersive platforms like Android XR. They drop the roles human hands play in conversation — pointing, describing shape, and emphasizing speech. ❓ How does the system actually generate gestures? 💡 An environment-perception module registers objects into a 3D spatial registry, and an LLM emits "GestureEvents" tied to trigger words in its responses. A local parser syncs TTS playback and animation using word-level timestamps, so co-speech gestures land exactly with the spoken words. ❓ Did it actually make a difference? 💡 In a 12-person within-subjects study covering orchid care and 3D printer tasks, participants found spatial grounding and locating objects significantly easier (p < 0.05), and following complex procedures got easier too. Gesture-paired visual effects for safety warnings stood out as especially effective. ❓ How did users describe the experience? 💡 One theme was feeling like "a partner is guiding you" — participants reported the experience shifting from tool-like search toward genuinely embodied conversation. #XR# #AIAgents#