I’ve been thinking about that tweet from Karpathy. The one where he describes a kind of coding that doesn’t really feel like coding anymore. You talk to a model, accept all the diffs, copy-paste errors back in, and things mostly work. You don’t fully understand the code. You don’t really need to.
That’s where vibe coding starts.
But where does it go?
The way I see it, vibe coding is just the first visible layer of a deeper shift. It’s not about style. It’s about detachment. From logic. From structure. From having to hold the whole system in your head.
And it’s spreading. Quietly.
You start with weekend projects. Then you use the same tools at work. At first it’s to save time. Then it’s because the old way feels slow. Then it’s because the old way doesn’t even make sense anymore—too many layers, too much abstraction, too many things the model is just better at handling.
Eventually, your job isn’t to build systems. It’s to shape them. You describe what you want. You run a few experiments. You monitor the outcomes. You intervene when it drifts. The skills look more like gardening than engineering.
And that shift won’t stop with code.
If “vibe coding” works, we’ll get vibe planning. Vibe research. Vibe operations. You’ll throw in a few prompts, feed some context, and see what comes back. If it’s not right, you tweak and try again. You don’t trace every line. You just trust the loop.
That’s not inherently bad. But it does mean something is changing. Fast.
Right now, most people still feel like they need to understand what’s happening under the hood. Soon, that might flip. Understanding could become optional. Or even a bottleneck. The faster route might be: prompt, review, deploy.
Which raises the bigger question.
What happens when most knowledge work moves to that mode? When more of the world is run by people giving input they barely understand, to systems they mostly don’t control?
We’re not there yet. But vibe coding is the shape of that future, just barely visible. Not the endpoint—just the first sign that the center of the work is starting to shift.
And once it shifts, we’ll have to decide whether to chase understanding—or let it go.