Three things run the AI world: data, hardware, and code. That's it. Everything else is downstream of those three.
Here's what people get wrong though - they think math is the foundation. It's not. Math is the explanation. You build the architecture first, you get it working, and then the math shows up to tell you why it worked. It's post-rationalisation dressed up in Greek letters.
Architecture is built for code. Always has been. The elegant theory comes after someone got their hands dirty and made something that actually ran.
This matters because there is too much gatekeeping AI behind (supposedly elegant) math before they've touched a single model. That's backwards. Completely backwards.
Build first. Break things. Get the architecture humming. The math will be waiting for you on the other side - and honestly, it'll make a lot more sense when it is.