A proactive agent needs to know when the work is finished, or I get a subscription to suggestions I never requested. The Gemini Spark announcement in May 2026 presented continuous tasks and proactive assistance. I like the idea of work moving forward while I do something else, as long as it is just as clear how to change direction and end it.
The problem is following progress. To understand where a task stopped, I do not want to read every conversation from the beginning. I need to know what already contributes to the deliverable and what is still missing. A list of executed commands leaves me reconstructing the history myself, as the context manager.
The same applies when several agents work in parallel and need to stay on the same request. Two runs can be working on the same question, or one can be waiting for the other, and whoever follows their progress needs to notice that before opening more work. Adding an agent to repeat the research just produces more answers for someone to assemble later.
Interruption has a cost too. A result that unlocks a decision deserves my attention. An intermediate detail can wait for the summary. I prefer adjusting that behavior around what I need to decide, instead of receiving a notification whenever the system shows it is alive.
My position: a proactive agent has to expose the objective, the stopping condition and who follows progress, and let me change any of them. Before it starts, it should be clear whether the request has a final deliverable or a reason for continuous monitoring. Without that, I am back to reading conversations to find out where the work stopped.