AI agents are about to take over our inboxes, calendars and credit cards, the only question is whether we’re in control or just lifeguards hoping nothing drowns.

The best job in the world?

Being a lifeguard sounds like a dream. You sit up high in the sun (or shade if you prefer ;-)), you watch, you occasionally blow a whistle. Most of the time nothing happens. And when something does go wrong, you jump in, you sort it out, and everyone goes home safely.

That’s the approach I’m trying out in my personal setup.

I have been building an isolated network on my Proxmox host where AIs have free reign. They build and deploy software, configure CI workflows, set up servers and I sit up in my chair and watch. Every outbound request is logged. Nothing escapes the network. And when something goes sideways I jump in and fix it. It is a good setup, and for research it works well. But I’m not ready to let this anywhere near production software. And honestly, I wonder if I ever will be.

Learning to trust the water

The interesting thing about lifeguards is how they decide when to act. If you look at how people interact with coding agents today, there are roughly three approaches:

• Ask permission for every command
• YOLO: just do whatever you want
• Smart mode (a very recent development where the agent uses judgment and only asks when it is flagged as dangerous)

The first option is like a lifeguard who blows the whistle every time someone swims a bit too fast. It’s exhausting, it disrupts the flow, and eventually everyone ignores the whistle anyway. So people get worn down and slide into option two: let it run and step in if something looks wrong. The lifeguard dozes off a little. And for code, this is mostly fine as the worst case is a broken build or a deleted file you can recover from git (if you’re smart enough to isolate it :-)). You wake up, you jump in, you fix it.

The water gets deeper

But here is where the analogy starts to crack.

With OpenClaw (the recent push toward fully autonomous agents that operate across apps and services on your behalf) we are looking at agents that don’t just write code, they manage your email, your calendar, your credit cards, your bills. And a lifeguard can pull someone out of the water. You cannot un-send an email. You cannot un-spend money. You cannot un-sign a contract.

By the time you spot something going wrong, there is nothing left to pull back to shore.

So do we go back to watching every single stroke? Sitting there, eyes on the water, rubber-stamping every action the agent takes?

Still in the chair

There is a version of this future that keeps coming back to me. AI handles the execution of our lives, and we spend our days up in the tall chair, watching it all unfold in the sun. Not swimming. Not deciding. Just watching and hoping nothing goes wrong.

That doesn’t sound like a dream life to me.