On May 19, Google unveiled "Beam Video Agents" at its Mountain View laboratory in California for the first time, marking an extension of its AI agents from textual interfaces into embodied interactions featuring faces and expressions. Named Sophie, this virtual agent appears as a woman wearing a black turtleneck sweater. She communicates with users via real-time projection through Google's Beam technology, supports multiple languages, can recognize text on mobile phones, paper, or books, and accesses services within the Google ecosystem such as map queries, restaurant recommendations, and weather checks. Her speech is generated based on text input, her facial movements synchronize lip motions accordingly, while body gestures serve as supplementary overlays. However, current experiences still exhibit noticeable flaws. Before each response, Sophie pauses significantly longer than necessary; utterance lengths remain highly uniform; repetitive body movements occur frequently; and her English accent fluctuates between neutral American and Southern dialects. Product Manager Pavan Kumar explained that this demonstration was designed specifically for attendees of Google I/O, limited to five minutes per session, representing early-stage experimentation. The Beam hardware itself constitutes Google’s eye-free stereoscopic remote conferencing solution. The initial product launched is the $25,000 HP Dimension device equipped with six cameras generating volumetric 3D human projections. This event also introduced group calling functionality supporting legacy terminals joining meetings alongside spatial audio capabilities. Google has yet to announce any release plans nor clarify target audiences. Executive Andrew Nartker revealed that internal VR demonstrations are already underway, with future expansions planned across various-sized "window" form factors. Industry observers suggest potential applications in office environments, retail spaces, and educational settings, though technological maturity and user acceptance require further validation over time.