This paper explores the challenges and opportunities of designing context-aware voice interaction for complex tasks, using cooking as a case study. The authors conducted two studies to evaluate how context-awareness in Voice Agents (VAs) affects user interaction. In the first study, they analyzed interactions with a commercial VA (Google Assistant) during cooking, identifying issues such as irrelevant responses, misinterpretation of requests, and information overload. These challenges were attributed to a lack of contextual awareness, leading to difficulties in maintaining shared understanding between users and the VA. In the second study, they evaluated interactions with a context-aware VA, where a human wizard simulated the VA's ability to understand the cooking task and provide relevant information. The results showed more fluent and complex interactions, with users making more explicit requests and using shared context to ground their conversations. The context-aware VA was able to provide proactive suggestions, reducing the need for users to repeat instructions and improving the overall interaction. The study highlights the importance of context-awareness in VAs for supporting complex tasks, as it allows for more natural and effective communication. The authors discuss the potential for personalization, the division of labor in VA communication, and the balance between proactivity and user agency. They also connect their findings to recent advances in generative models and multi-modal machine learning approaches to conversational interaction. The paper concludes that context-aware VAs can significantly improve user interaction by enabling more natural, fluid, and effective communication. This is particularly important for complex tasks like cooking, where shared context and proactive support are crucial for successful task completion. The study provides valuable insights into the design of context-aware VAs and their potential to support users in complex, multi-stage tasks.This paper explores the challenges and opportunities of designing context-aware voice interaction for complex tasks, using cooking as a case study. The authors conducted two studies to evaluate how context-awareness in Voice Agents (VAs) affects user interaction. In the first study, they analyzed interactions with a commercial VA (Google Assistant) during cooking, identifying issues such as irrelevant responses, misinterpretation of requests, and information overload. These challenges were attributed to a lack of contextual awareness, leading to difficulties in maintaining shared understanding between users and the VA. In the second study, they evaluated interactions with a context-aware VA, where a human wizard simulated the VA's ability to understand the cooking task and provide relevant information. The results showed more fluent and complex interactions, with users making more explicit requests and using shared context to ground their conversations. The context-aware VA was able to provide proactive suggestions, reducing the need for users to repeat instructions and improving the overall interaction. The study highlights the importance of context-awareness in VAs for supporting complex tasks, as it allows for more natural and effective communication. The authors discuss the potential for personalization, the division of labor in VA communication, and the balance between proactivity and user agency. They also connect their findings to recent advances in generative models and multi-modal machine learning approaches to conversational interaction. The paper concludes that context-aware VAs can significantly improve user interaction by enabling more natural, fluid, and effective communication. This is particularly important for complex tasks like cooking, where shared context and proactive support are crucial for successful task completion. The study provides valuable insights into the design of context-aware VAs and their potential to support users in complex, multi-stage tasks.