Fire and Forget: Building Asynchronous, Long-Running AI Agents with Google ADK Google's Agent Development Kit (ADK) supports asynchronous, long-running AI agents that can pause, resume, and incorporate human-in-the-loop steps, according to a technical write-up on building such agents. The piece argues most real agents run for minutes, hours, or days rather than following a quick conversational question-and-answer pattern, so communication with humans must be either synchronous or asynchronous. It covers pausing and resuming agents and long-running operations in Google ADK. Pausing and resuming agents, human-in-the-loop, and long-running operations in Google ADK In most of agent use cases today there is a UI in which the user and the agent have a conversation. This works for certain use cases where questions and answers are quick, but I believe most agents simply don’t follow a “conversational” pattern. Because let’s be honest: real agents today have a long lifetime. They can span minutes, hours, maybe days. And obviously, you don’t wait for hours in front of your chat window. Communication with humans can follow two patterns: synchronous and asynchronous.