Do not go to sleep until you have started at least one long-running task.
I mean work that can continue for a day or more. Preparing that task makes you think beyond the next reply. What do you want to have when it finishes? What steps can happen without you? How will you know the result is any good?
With an agent beside you, it is easy to answer those questions one message at a time. Leaving it work to do while you are away means answering more of them upfront. You need to give it a direction, enough context to proceed, and a way to check its work.
For example, ask an agent to prepare and start a 30-hour stability test for a service in a test environment. Long tests can reveal gradual performance degradation and resource leaks that shorter runs miss (Grafana's explanation).
Decide the load, acceptable response times, and resource limits before starting. Run a short check first, then have the test save measurements and failures throughout the run and stop if it crosses the limits. Ask the agent to return the raw results with its report so you can check the conclusion.
The test does the waiting here. The agent helps prepare and inspect it. You still have to decide what a useful result looks like before committing a day to it.
Do not invent work just to keep something running. Pick something whose result will help you decide what to do next. The habit I am arguing for is thinking far enough ahead to set that work in motion before you go to bed.