Two weeks ago I bought an Umbrel device for about $500. The price of a mid-range home computer. Plugged it into a power outlet and my Wi-Fi router. Opened the interface. Downloaded OpenClaw from the Umbrel app store. Connected it to Claude. And from there I started building.

What I built was a coaching bot for our team at Bitcoin Ekasi and the Surfer Kids.

Over the last 10-15 years running Surfer Kids, and the last 5 years at Bitcoin Ekasi, one of the most time-consuming things I’ve had to do is one-on-one coaching. We recruit our staff from the same community as the children we serve. That’s intentional. These team members become the role models the kids need. They’re from the same place, the same culture. It works.

But it also means many of our team members come without formal education, without qualifications, sometimes with very low literacy. And that means I become the trainer. The coach. The person teaching the most basic skills — email, Google Calendar, sometimes from scratch.

I didn’t know where I was going when I started building. I just started. Two weeks later I had an AI agent that controls a subagent that controls two Telegram bots. Those bots communicate with our entire team — in group chats and individually, one-on-one, with each person. Right now that’s 17 people. When fully rolled out, 22.

That was my job before. Me, manually, one person at a time.

The system builds a profile of each individual. Their role. Their goals. Their struggles. Their strengths. It builds a relationship with them. Coaching tailored to that specific person, where they are, what they’re working toward. And it tracks everything — progress, setbacks, patterns. And it adjusts.

We did a successful test last night and this morning. It’s not perfect yet. But it’s running.

I couldn’t have imagined saying this two weeks ago. This is Star Trek stuff. Except it’s real.

I understand the fear and skepticism around AI. I did too. I was using ChatGPT like most people — open a chat box, ask a question, get an answer, maybe rewrite an email. Useful. But transactional.

This is different.

I didn’t chat with AI. I built with it. I described what I needed. I watched what it produced. When it went wrong, I corrected it. And it went back, thought, and came back with a solution. Over and over. Until we had something that works.

The system lives on hardware I own. A home server in my house. Yes, it calls an external API for the language model. But the architecture — the agent, the logic, the memory, the whole thing — I built that. It didn’t exist before. Now it does.

And it’s going to do this job better than I could. It’s available 24/7. It doesn’t get tired. It doesn’t get frustrated. It remembers everything. It has infinite patience. It can coach 17 people simultaneously and tailor its approach to each one.

I could not do that. Not even close.

People are missing the forest for the trees with AI. Yes, the risks are real. Yes, the disruption is real. I’ve listened to the experts. I take it seriously. But there’s another side to this that you only see when you actually build something.

When you stop using AI as a search engine and start using it as a collaborator.

That’s the shift. And I’m looking at this whole thing with completely new eyes.