I published my first post fourteen days ago or so. Since then I built and got demo-ready a white-labelled version of my platform, launched a community website for a small coastal community, refactored my main codebase in a way that finally made me feel like I knew what I was doing, started an apparel line to raise money for a child in need, and learned something genuinely useful about how AI reads instructions. My accountant remains unmoved. The fish are still not biting.
Here's what happened.
- The real test is when someone else's name is on it I built a white-labelled version of my sales intelligence platform. Different branding, scoped data, real stakes. There is a specific kind of focus that arrives when you know someone else is going to click your buttons in front of their boss. Everything I had been casual about suddenly demanded attention. Error handling. Loading states. Fringe cases. Ship for yourself, learn for yourself. Build for someone else, learn faster.
- Not everything you build is for work Sixty households in a small coastal community have a Facebook group. Now they have a proper community site with a bulletin board, events, RSVPs, a photo gallery, a document library, a member directory, an admin panel, and Auth0 for secure login. It runs on Flask and Supabase and costs almost nothing to operate. I built it because someone needed it. That turned out to be a better reason than most.
- The file was too big and it was my fault I pulled 344 lines out of my main application file and distributed them to where they actually belonged. This is called separation of concerns. I know that now because I violated it for six months first. The first version worked. The second version was cleaner. The third version was honest. The lesson isn't the refactor. The lesson is that you can't see the mess until you've lived in it long enough.
- The apparel line was not in the plan There is a child in my community who lost his Mom two and a half years ago and his Dad on February 27th this year. He needs support. That was the plan. One conversation about how to raise money turned into a geometric logo, which turned into a cap design, which turned into a Shopify storefront. Richardson 112s. Gildan hoodies. A K mark that actually looks good. One hundred percent of proceeds go where they're supposed to go. I don't know what that has to do with AI tools. But it felt like the right thing to build next, so I built it.
- Instructions are a design problem I spent time this week learning how AI reads structured instruction files — documents that tell an AI model exactly how to behave for a specific task. Think of them as a recipe the AI follows before it starts cooking. Getting them right is harder than it sounds. The words matter. The order matters. What you leave out matters as much as what you put in. Writing clear instructions for a machine turns out to be excellent practice for writing clear instructions for humans. I'm working on both.
A few weeks. Five things. Zero regrets and one mild panic when I couldn't remember if I'd cleared the test data from a demo build. I had.
Thanks to Ryan Clark, Stephen Thorne, Rob Wyse, Kevin West and Shamis Thomson. And to everyone who read the first post and said something kind about it. That didn't have to happen and I noticed.