What do you do when your AI agent needs you and you're not at your desk?

Coding agents can now work for hours on their own. But I kept running into two dumb problems:

1. The Mac falls asleep. I'd give Claude Code a big refactor, close my MacBook to head out, and come back to a dead session.

2. The agent waits for me. If I stayed, I'd spend the evening at my desk just typing "yes" every time it asked "can I run this?". If I left, it sat there waiting for hours.

I tried caffeinate, Amphetamine and leaving the lid open with the brightness down. None of them knew whether an agent was actually working, and none of them could answer the agent when I wasn't there.

So I built Latte: a small menu-bar app that keeps the Mac awake (even lid-closed) only while agents are working, sends their approvals to my phone through Telegram, and lets the Mac sleep once the work is done. It's launching here today: [PH link]

But I'm more curious how others handle this:

- Do you leave agents running when you step away, or do you wait it out?

- Have you lost a long run to sleep, a closed lid or an unanswered prompt?

- What's your current setup: tmux, a remote box, a spare Mac, just staying at your desk?

Would love to hear your stories (and the hacks).