The first time I had a real conversation with ChatGPT, I did what any self-respecting engineer would do — I tried to break it. I threw ambiguous sentences at it, asked it to explain recursion using only cooking metaphors, had it write code in languages I made up on the spot. And it just… worked.

I’d been following transformer models since 2017. I understood the theory, had even built smaller versions myself. At Codika, where we’re building systems that transform business requirements into running software, I spend a lot of time thinking about this gap between mechanism and capability. We know exactly how transformers work. But there’s a massive gap between understanding how something works on paper and seeing it actually hold a coherent conversation about, well, anything. The architecture I understood, but the emergent behaviors? Those were something else entirely.

That’s what this series is about. Not explaining transformers to ML engineers, you already have plenty of resources for that. This is about making these concepts accessible to everyone else: the developers who use these models but haven’t dug into the architecture, the product managers trying to understand what’s possible, the curious minds who want to peek under the hood without drowning in mathematics.

Let’s start at the beginning, not with transformers themselves, but with the problem

…

The first time I had a real conversation with ChatGPT, I did what any self-respecting engineer would do — I tried to break it. I threw ambiguous sentences at it, asked it to explain recursion using only cooking metaphors, had it write code in languages I made up on the spot. And it just… worked.

I’d been following transformer models since 2017. I understood the theory, had even built smaller versions myself. At Codika, where we’re building systems that transform business requirements into running software, I spend a lot of time thinking about this gap between mechanism and capability. We know exactly how transformers work. But there’s a massive gap between understanding how something works on paper and seeing it actually hold a coherent conversation about, well, anything. The architecture I understood, but the emergent behaviors? Those were something else entirely.

That’s what this series is about. Not explaining transformers to ML engineers, you already have plenty of resources for that. This is about making these concepts accessible to everyone else: the developers who use these models but haven’t dug into the architecture, the product managers trying to understand what’s possible, the curious minds who want to peek under the hood without drowning in mathematics.

Let’s start at the beginning, not with transformers themselves, but with the problem

…