Overhearing AI Agents represent a new direction in AI assistance, shifting focus from individual user support to passive, multi-human context awareness. Unlike traditional AI copilots that monitor a single user’s isolated actions—such as writing code or drafting text—overhearing agents listen in on group interactions, capturing ambient, multi-modal data like spoken conversations, gestures, or shared activities. The core idea is to infer collective intentions from these interactions without joining the conversation, enabling subtle, timely interventions like pulling up relevant case histories or generating visual aids such as flow diagrams. This paradigm is distinct from the co-pilot model not because of its passivity, but because it operates within multi-user environments where direct AI participation could disrupt natural interaction. In a team meeting, a family discussion, or a collaborative workspace, an overhearing agent observes without speaking, learning from context and timing its help to be most useful. While the framework can extend to single-user scenarios—such as monitoring a person writing a document or coding—it is designed primarily for group settings where the value lies in understanding shared dynamics. The paper situates overhearing agents within a broader taxonomy of AI agent approaches. Conversational agents engage users directly through back-and-forth dialogue, relying on LLMs to break down complex tasks and plan responses. Autonomous agents go further by generating detailed plans or sequences of actions based on user goals, often with minimal input, and allow users to review and approve them before execution. These agents simulate world models and can initiate proactive actions based on environmental cues. In contrast, the co-pilot approach focuses on real-time augmentation during solo work, offering suggestions without interrupting the user’s flow. It’s predictive and context-aware but limited to one person’s activity. Overhearing agents expand on this by capturing the broader social and environmental context, allowing the AI to understand not just what one person is doing, but how multiple people are interacting and what shared goals might be emerging. This shift reflects a deeper evolution in how AI integrates into human environments—not as a participant, but as a silent observer that enhances collaboration. By processing audio, video, and other ambient signals, overhearing agents can detect patterns in group behavior, anticipate needs, and offer support that feels natural and timely. The challenge lies in balancing privacy, accuracy, and relevance, but the potential for improving teamwork, decision-making, and creativity in complex settings is significant.