How to Stop Babysitting Your AI Agent

You tell Claude Code to refactor the authentication module. It says it'll take a while. You stand up, walk to the kitchen, start making coffee — and immediately start thinking about whether it's finished yet.

Two minutes later you're back at your desk. It's still running. You walk away again. Ninety seconds later, you're back. Still running.

You try to do some cleaning. You fold one shirt and think: maybe it's done now. You go check. It's not. You try again. This time it finished three minutes ago and has been sitting there waiting for you to tell it what to do next.

If you use Claude Code or Codex and work from home, you know this exact loop.

The anxiety is rational

This isn't a discipline problem. The anxiety makes perfect sense, because there is genuinely no good way to know when your agent is done.

Claude Code has no native notification system. There's no push notification to your phone. There's no sound that carries beyond your desk. You can set a terminal bell, but it's a quiet blip — completely inaudible from the next room, let alone while running the vacuum or making coffee.

So the only way to know when it's done is to come back and look. And since you don't know when to come back, you either check constantly or accept that your agent might sit idle for 10, 20, 30 minutes before you notice.

Neither option is great. Checking every two minutes means you never truly step away. Accepting the delay means wasting time that compounds across every task, every day.

What you actually need

The problem isn't complicated. You need one thing: a solid, unmissable signal that says "the job is done, come back to your desk."

Not a silent badge. Not a tiny terminal beep. Not a chat message in a tab you're not looking at. You need something that reaches you in the kitchen, in the garden, in the shower — wherever you went when you walked away.

A phone call does exactly that. Your phone rings. You hear it from anywhere in the house. You know immediately: the task is done, the agent needs you, it's time to go back.

That one signal eliminates the entire anxiety loop. You stop checking because you know you'll be told. You stop half-doing other things because you can fully commit to them. The mental load of "is it done yet?" just goes away.

What the loop looks like with a phone call

Here's the same scenario, but with AgentPing:

1. You tell Claude Code: "Refactor the auth module. Call me when it's done."
2. You walk to the kitchen. You make coffee. You actually drink it.
3. Fifteen minutes later, your phone rings.
4. You walk back to your desk, review the results, and kick off the next step.

No checking. No anxiety. No wasted idle time. The agent calls you when it's ready, and you come back exactly when you're needed.

And if the agent gets stuck partway through — needs a decision, hits an error, can't resolve an ambiguity — it calls you then too. You don't come back to a stale terminal that's been silently waiting for 20 minutes.

Four patterns that replace the checking habit

"Call me when it's done"

The most common one. You start a long task and walk away. When the agent finishes, your phone rings. Simple.

"Call me if you get stuck"

You want the agent to keep working autonomously, but call you if it hits something it can't resolve. Your phone only rings if there's a real blocker — otherwise, silence.

"Call me if I don't reply in 10 minutes"

The agent posts results in chat first. If you happen to be at your desk and reply, no call. If you've walked away and don't respond within 10 minutes, then your phone rings as a fallback. Chat stays primary; the call is the safety net.

"Call me if it fails"

You kick off a deploy or test suite. If everything passes — silence. If something breaks, immediate call. You only get interrupted when it actually matters.

The deeper benefit: you stop hovering

Here's the part that surprised us. Once you trust that you'll get a call, you don't just stop checking — you start giving your agent harder and longer tasks.

Before, you'd break work into small steps because you didn't want to walk away from a 30-minute task. Now you can kick off a full migration, go for a walk, and come back when it's done.

The safety net doesn't just save you time. It changes how you use your agent. You give it more autonomy because you trust the escalation path. And more autonomy means more leverage from the same tool.

Getting started

1. Sign up for a free account and verify your phone number
2. Install the AgentPing skill in Claude Code or OpenClaw, or use the REST API from any agent
3. Next time you start a long task, say "call me when it's done" and actually walk away

The free plan includes 10 alerts per month and 5 call minutes — enough to see if it changes how you work.

The goal isn't to get more phone calls. It's to stop checking your terminal every two minutes — and finally let your agent do its job while you do yours.