Start Smaller Than I Did

If you are around my age, not technical, and thinking about building something with AI, here is the advice I wish I had heard before I started Beads of Hope T1D.

Start smaller than I did.

On June 2, I gave the AI one giant prompt. I asked for a full ecommerce and community site for Type 1 Diabetes awareness, with handmade beaded products, checkout, accounts, community features — basically the whole dream in one breath.

And to be fair, it did a lot. In about 20 minutes it created a shop page, Google login, Stripe checkout, and 22 database tables. I remember thinking, “Well, this is either magic or I’ve made a terrible mistake.”

Both feelings were probably correct.

What I would tell you is this: don’t begin with the whole house. Begin with the front door.

For me, that could have meant: “Make a simple shop page with four products and visible photos.” That alone would have saved me some gray hair, and at 50, I am not exactly looking to add more.

Because the first real wall I hit was not some fancy business problem. It was product images. They were invisible on the published site. When I clicked where the image should have been, the page snapped back to the top of the “all” category. Duplicate category buttons kept appearing too.

I typed things like, “images are still not visible and each time I click on the box the image should be in, the screen immediately reverts back to the top.”

Then I watched the AI try. And fail. And try again. And fail again.

For hours.

So my second piece of advice is: when something breaks, describe what you see like you are talking to a very patient repair person on the phone. Don’t worry about sounding technical. I didn’t know the right words. I just kept explaining what was happening on the screen.

At one point I got an error that said, “Cannot find module '@/assets/blue-ribbon.png.asset.json'.” I had no idea what an asset.json was. None. It sounded like something I was supposed to pretend to understand.

Instead, I asked. The AI explained that it was basically a pointer file for images, and eventually regenerated it.

That taught me something important: you do not have to understand every word to keep going. But you do have to be willing to ask the “embarrassing” questions.

Also, be prepared for the tiny details to take over your life for a while.

I spent a Sunday asking the AI to move a blue ribbon graphic down a quarter inch, put the heart closer to the ribbon crossover, add white shading behind “T1D Hope,” and adjust a hover-zoom on badge reel images. First I wanted 3x zoom. Then 2x. Then only on badge reels, not everything.

Would 99% of visitors notice? Probably not.

Did I care deeply at the time? Absolutely.

That is another thing I’d tell you: your project may look like “just a website” to someone else, but if it is connected to someone you love, it will not feel small. This one is for my son, who was diagnosed with Type 1 Diabetes when he was two. That made every ribbon, every word, every photo feel personal.

But personal does not mean you should skip the serious parts.

When security scans flagged things like “guest cart hijacking,” “client-controlled unit price,” and “RLS policies too loose” (database rules that decide who can read or change what), I understood maybe 40% of it. Maybe.

Still, I didn’t ignore it. I had the AI fix the issues and run the scan again. If money or customer information is involved, do not just assume the AI got it right the first time.

And speaking of money, know that the emotional moment may not be when the site appears. For me, it was Stripe. Adding bank information, business details, two-factor login, thinking about shipping and tax — that was when I thought, “Am I really doing this?”

That part felt more real than the code.

So if you are starting today, I would say: build one piece first. Test it. Click everything. Ask plain questions. Don’t be ashamed when you don’t understand the error message. And when the AI gives you something too big, tell it to slow down.

I’m still nervous about the first real order. The store is live-ready, but Stripe activation is still sitting there like a final deep breath before jumping in.

What I learned: the AI can build fast, but I have to steer slowly.