When should an AI admin assistant ask before acting on something it noticed?

An assistant that handles travel, email and scheduling is useful partly because it sees the loose threads people forget. Catch says its admin AI works through Slack, Messages, email and calls, adapts to priorities, and spots problems early. The awkward case is the thing it notices but was never given. A meeting note mentions a child’s appointment. A forwarded thread reveals a budget worry. A calendar gap looks like an opening. “Helpful” can turn into a new obligation before the person has decided it is work. I would want the assistant to distinguish “you asked me to handle this” from “I inferred this might matter.” The first can move. The second should arrive as a small question, with no calendar hold, message, or follow-up already underway. What should stay in the suggestion pile even when an AI admin assistant is trying to save you time?