Beyond Codex and Claude: what do you use agents for?

I’ve been experimenting with agents like Hermes and OpenClaw, and the most interesting use cases I’ve found aren’t really about chatting or coding. They’re tasks that continue after I close my laptop: checking something periodically, maintaining a server, watching for changes, or taking action only when certain conditions are met. That feels quite different from how I use Codex or Claude. With those tools, I’m usually present and actively driving the session. With an always-on agent, I’m delegating something and coming back later. I’m curious where others have found this distinction actually useful. What have you delegated to an agent that you wouldn’t naturally do in a coding-agent session?