After the phone and the laptop came Raspberry Pis.
My first real project was Hawke — an open-source tracker. I opened ports on my router to get access off the LAN, set up a reverse proxy. It worked. Very badly. I quickly realised that exposing ports to the internet was a terrible idea, and moved to Nextcloud for photo uploads on the local network only. It worked OK but I honestly didn’t understand why.
For a while I settled on using Pis as desktops — one for me, one for the wife. Wasn’t great, but it kind of worked. Set up a Bitcoin node. Again, worked badly — no incoming connections. There was a pattern: try something, break it, learn one thing, try again.
Then I hit a wall. I wanted to build a family notification service using Signal-CLI — weather updates, travel info, that kind of thing. Signal-CLI is not user-friendly. I was completely stuck.
I’d been using AI tools around this time, but only like Google — asking questions, reading answers. Then ChatGPT said something that changed everything: “I can write the code for you.” And it did. Before long I had a running Signal-CLI server giving updates to family members.
That was the moment the ceiling lifted. Not because AI wrote perfect code — it didn’t. But because I could now attempt things I would never have attempted alone. The cycle time from idea to something running collapsed completely.