Dostoevsky. A burner phone. Pi.

This last weekend, my 17-year-old son Jared and I built a fully functioning AI agent on a $125 Raspberry Pi. Connected to email (Fyodor Dostoevsky's email, rather). Executing real tasks. On a computer smaller than a paperback. Jared was locked in. I was locked in. It was one of those rare Saturdays that just disappears. And then the enterprise brain kicked in. Somewhere between getting the agent running and watching Jared lose his mind with excitement, I started mentally drafting a security review. Credentialing. Access controls. What happens when this thing has real permissions? How do you write a runbook for an agent you built in an afternoon? I got so turned around that I bought a burner phone to connect it to instead of my real email. My son thought I was insane. He was probably right. But here's what I kept thinking about: my customers are already doing this. Maybe not on a Pi, maybe not with a 17-year-old cheering them on — but the scrappiness, the speed, the "let's just try it and see" energy? That's real. And it's happening faster than most enterprise roadmaps account for. The questions I was asking on my kitchen table — security, governance, how do you operationalize something that moved this fast — are the exact same ones I hear from customers every week. The gap between "we built something cool" and "we can run this in production" is where the real work lives. Jared doesn't lose sleep over that gap yet. Someday he will. For now I'm just glad one of us was having fun. What's the scrappiest AI build you've seen actually make it to production?