How DeepAgents could handle asynchronous subagent updates

The post’s author says they’re considering how to make DeepAgents more async-native. For a finished subagent, they outline two options while the parent agent is running: interrupt it and continue after applying the update, or wait until its run ends and batch the update with other messages. If the parent is paused for a human response, they ask whether the update should wait too or resume the thread, cancelling the pause and potentially prompting for approval again.