Adventures with Claude

I’ve never understood the appeal of AI. I’ve just never really “gotten” it. I’ve seen the viral chatgpt art, like the action figures and studio Ghibli trends, and I have friends who can’t go an hour without asking “their friend chat” or copilot a question. But I didn’t see the need for any of it. That is until I discovered vibecoding.

I stumbled into it without really knowing what I was doing at all. I felt like a pioneer, even though I’ve since discovered it’s fairly common. it’s been a few weeks and I have now built over 30 apps, ranging from silly personal projects, to down right complex docker containers that interconnect to make life easier

It started with a simple request. Our business website had a script that pulled enrolment classes from our class management system (Momence), that we had paid a web designer to create for us. But this guy had fallen off the face of the earth and become completely uncontactable. But it wasn’t working quite right. We wanted classes to stick around on the site for at least three weeks to allow students to enrol late, but no matter how much I dug into the code, I just couldn’t figure it out. I’m by no means a technical expert, but I get by. I know yaml, and some basic css, but this was next level, with hundreds of lines of api calls and filters.

I decided to ask ChatGPT. It was useless. Less than. It kept telling me it had done it, when it hadn’t. It would just put emojis in to the script, and not fix the code. I tried Gemini. Even worse. I googled, I tested. By the end of the week I was up to semesters39.js. It still wasn’t working.

I turned to Claude. It was immediately clear that Claude was different. It felt like a partnership. Claude would try something, then bounce ideas off me. It would give me advice on how to look into the source api in my browser devtools. Eventually we found it. The api itself was dropping classes. There was no way to get it to do what I wanted.

Then Claude had an idea.

”We‘ll cache the api into a json. A python script that runs nightly. Retains classes for three weeks. Host it on your server, point js at that instead of base api.“

And you know what? It worked. And Better than worked, it’s still working. It even displays when a class started.

Claude is like my own personal Rocky (If you haven’t seen or read Project Hail Mary yet, go do it now). It’s like I’m collaborating with the smartest entity on the planet. Does it get everything right? Not by a long shot. But together, we can figure it out.

I felt unstoppable. What else could I build. The list in my head grew large.

I love movies. You’ll know that if you’ve listened to any of my podcasts (shameless Poddy Mouth Radio plug here), and I try get to the movies at least once a week. But I am not discriminative about which cinema I go to. There are 5 in my town, and no aggregate source to see movie times for all of them. So I asked Claude if we could plug it into my home assistant. This time I came prepared. I dug into the devtools as I browses movie times on each site. I made curl requests to check I had found what I thought I had. Within an hour, we had a python script and yaml file that would scrape all movie show times for today and tomorrow, combine by movie title, and present in home assistant.

I could even set up messenger notifications, every Monday, tell me what’s on at the Southside cinema between 3:30 and 4:20pm, so I know what movies I can catch while my daughter is at drama. I know this sounds like a stupid build. But it changed my life.

Over the next few weeks I’ll write a few articles here covering all my different builds. The trials and errors, the roaring successes.

And you know what - I am not using AI to write any of this.