Vibe coding is like magic in Frieren
Visualization = Realization
It randomly occurred to me recently that there are a lot of parallels between modern software engineering with coding agents and the magic system in Frieren. Not sure if anybody else has talked about it but I’ll just dump my thoughts here.
In Frieren, there’s a scene where Ubel talks about being able to cut anything she is able to visualize, but not being able to cut anything she isn’t able to visualize. To me, high-quality visualization means not just simply “imagining that something happens”, but rather imagining in extremely granular detail what something should look like if it were to happen. I imagine the skill of a mage in the Frieren world would be correlated with their exact ability to visualize how spells get realized. In other words, the more detail in which you can visualize something, the stronger your ability to realize it.
Coding agents’ ability to realize is determined by the user’s ability to visualize
There is a weird parallel of this visualization = realization idea to writing code with a coding agent. Everybody has access to coding agents nowadays, but the impact delivered by people using them is highly variable. There are cases where coding agents amplify impact 2x, 10x, 100x, yet also at the same time there are cases where the generated code is complete vibe-coded slop. There also seems to be highly varying opinions about the capabilities of coding agents, where some people claim the agents are dumb and completely unable to do anything useful, and others who claim they haven’t manually written a line of code in months.
I’d like to throw out a hypothesis here: the impact of a software engineer with a coding agent is determined directly by their ability to visualize what they’re building. The more detail in which you can visualize what you want to build, the higher the probability you’ll be able to realize that exact thing you wanted, instead of getting just slop code.
It is probably actually impossible for AI to one-shot the exact thing you wanted to build if you don’t specify it in detail, because the space of things that fit your description is too large otherwise.
The ability to visualize has always been important, even before coding agents and AI, but it feels like it’s becoming especially important now as coding agents get better. Prior to coding agents, maybe you could visualize how something works, but it might have taken massive effort to learn all the coding frameworks, concepts, libraries, etc to achieve that. However, as coding agents get better, the need for prerequisite knowledge approaches zero, and instead the ability to visualize what is happening and what you want becomes the determing factor for impact.
Coding agents amplify experts’ impact more because they can visualize better
Naturally, deeper knowledge in a field is correlated with more detailed visualization, which is why I think it makes total sense that an expert with a coding agent will always be able to deliver more impact than a novice with a coding agent.
Novices may be generating non-optimal, inaccurate, slop code without even realizing what they are producing is slop. That’s not actually a limitation of AI, but rather a limitation of the user. Since the novice is actually unable to distinguish between what’s slop vs not slop (good vs bad), it is actually totally possible that the coding agent may generate something that perfectly satisfies the user’s description, yet is simultaneously bad.
A final conclusion we can draw from this is that in today’s state of AI and coding agents, it still looks like AI won’t be able to replace humans yet - and actually it’s quite the opposite. Experts are actually able to increasingly amplify their positive impact more than ever, while novices are only able to amplify generation of slop (negative impact) since they cannot actually visualize the difference between slop vs not slop.