The Case Against Magic
Technology may feel like magic, but we need to remember it isn't.
There seems to be a paradigm where we associate vibe coding with sloppy and low-quality outputs. After all, it gets its name from the fact that it’s based on ~vibes~ and not specifics. However, vibe coding has armed people like me with the ability to create outputs in a medium that wasn’t possible for us before - at least not without the investment of learning to code by hand. It’s become so easy to write a program that some might even say it feels like magic.
This morning, a tweet from my friend Ivan (co-founder of the AI company Cohere) came up on my timeline:
(Technically, this is Nick Frosst’s quote)
You can’t think the technology is magic.
My first reaction was “Why not?” What we consider well-designed technology usually carries a sense of mysticism. The times I’ve been most impressed by technology were ones where I’ve thought, “How did it just know?”
Then I considered where this statement is coming from. Cohere develops technology, and to develop technology, you need to understand technology. Having a deep understanding will inevitably clear the fog and dispel the mysticism. Magic can feel like a shot in the dark or a roll of the dice, but to make something groundbreaking, one must be precise and know what they are doing.
This makes me think about some of the conversations I’ve had about my projects recently. When people ask how I built them, my response is usually a credit to Codex/Claude, and the reaction I get is often “Really? Is that all?”
I think there is still a high degree of mysticism around how certain AI-generated outputs are created, and while I acknowledge that LLMs are black boxes, the ways we can wield them are more similar to discovering new land than conjuring portals. While a well-designed UI may have been created using AI, conversations with designers like Liam from Browserbase reveal that good design still follows frameworks such as font selection and consistent alignments (not magic). Charting this territory looks less like stringing together the perfect description of vibes to throw into the black box, and more like a guide of discovered or developed functionalities and how to apply them (i.e., this guide).
Likewise, AI may be able to execute a novel idea, but the idea itself is still the distillation of the creator’s references and experiences of our world. To an outsider, how another person comes up with something they consider creative and unique might seem like magic, but to the creator, the strings that tie everything together are very clear. The models will eventually learn many of these frameworks as we continue to document them, but my point is that the reason why something looks good or feels unique isn’t magic, it’s a principle that someone had to figure out.
I’m not sure what the most common definition of “vibe coding” is right now, but I like to think of it as a neutral executional process or a tool rather than something that carries a quality label (I understand why we associate vibe coding with low quality outputs since more amateurs making things will inevitably result in more low quality results, but I think that is more of a symptom than an inherent trait). As coding agents make it possible for anyone to express an idea through technology (software + hardware), I’m optimistic about a future where we have dedicated creative tools like Adobe suite, DAWs, and Figma for this growing format of expression. I place my bets on the ones rooted in process, understanding, and precision - not the ones trying to sell magic.