I think most developers have had the same experience with AI coding assistants. You give the model a vague instruction, it produces something that looks suspiciously impressive and for a few minutes you feel like you’ve figured out the future of programming. Then something breaks.
And then you realise you have absolutely no idea why.
That is pretty much where Process Over Magic begins.
Uberto Barbini isn’t against AI-assisted coding. Quite the opposite. He just seems deeply unimpressed by the idea that we can throw a clever prompt at an LLM and call it software engineering. The book’s argument is simple. AI can make us faster, but only if the process around it is good enough to keep that speed from becoming chaos.
I liked that the book doesn’t start by pretending vibe coding is useless. Barbini actually lets it happen. He builds a game, shows what happens when you let the AI run with an idea and then starts pulling apart why the result isn’t enough. That made the argument much more convincing than simply telling developers to “be disciplined.” You get to see the difference.
Get Process Over Magic by Uberto Barbini Here!
From there, the book gets into specifications, commits, testing, modular design and refactoring. None of these are particularly new ideas, and that’s actually the point. The interesting part is what happens when you apply those existing engineering habits to working with an AI assistant. The machine changes the speed of the work. It doesn’t remove the need for knowing what good work looks like.
The “one prompt, one commit” approach particularly stood out to me. It sounds almost overly cautious until you think about how quickly an AI assistant can change a codebase. Keeping changes small and traceable makes a lot more sense when the person producing the code isn’t necessarily the person who understands every line of it.
I also appreciated that Barbini doesn’t stop at the shiny new project. The chapters on large and legacy codebases are probably more valuable for experienced developers because that’s where the fantasy of effortless AI coding starts to fall apart. Existing systems have history. They have technical debt. They have decisions nobody remembers making. AI doesn’t magically know any of that.
The book is fairly lean, which works in its favour. It doesn’t need hundreds of pages to convince you that process matters. The examples do most of the work.
If you’ve already played around with AI coding tools and you’re wondering what comes after the initial excitement, Process Over Magic is a good place to look. It doesn’t promise that AI will make you a better developer.
It asks you to become a better developer first. Then use AI to get faster.