Harrison Chase explores how ambient agents differ from conventional chatbots by reacting to real-time event streams instead of relying on explicit commands. These agents can handle multiple tasks simultaneously and require low latency for optimal performance. While they have the potential to scale impact significantly, human oversight is critical. Four interaction patterns are highlighted: approving or rejecting actions, editing actions, responding to inquiries, and utilizing 'time travel'. Human involvement ensures better outcomes and trust. Tools such as LangChain's agent inbox, LangGraph, and Langsmith provide insights and enhance the orchestration of these agents. Langsmith specifically supports long-running agents with visibility features, including an open-source email agent that drafts responses and sends calendar invites while maintaining human oversight.