I Spent the Weekend Actually Using AI Agents

Everyone’s debating whether AI agents are safe. I spent the weekend actually using OpenClaw.

Moltbook leaked 1.5M API keys. Agent security is a real and growing problem. The takes are valid. But the discourse has become so focused on what could go wrong that we’re missing what’s already going right.

I set up an agent on my spare Mac this weekend. Within 48 hours it had:

Researched 39 daycares for my kid (emails, reviews, Google Sheet)
Helped a friend complete their online traffic school
Drafted a hotel negotiation for my upcoming trip
Worked through tasks asynchronously while I was AFK

The combination of browser and CLI access is magical. What used to take a team of people doing grunt work is now a conversation away.

But the part that actually blew my mind was the self-improvement. The agent needed a Google Places integration it didn’t have. So it found the skill, installed it, and kept going. PDF parsing? Same thing. Like watching someone learn new abilities in real time.

There’s a scene in The Matrix where Trinity downloads the ability to fly a helicopter in seconds. “Can you fly that thing?” Three seconds later, she’s at the controls.

Trinity downloading helicopter flying skills in The Matrix

That’s not science fiction anymore. That’s my laptop.

The real story with agents isn’t the hype or the security drama. It’s that the productivity gain is so large it’s hard to explain without sounding like you’re exaggerating. Tasks that took hours happen in minutes. Research that needed a team happens while you sleep.

We’re living in the future. It just looks like a guy talking to his laptop.