i get the appeal of vibe coding, but in practice it often shifts the hard work instead of removing it. you write less code, but you spend more cycles reconciling what the AI produced with how the system actually works. for small experiments it’s great, but once a product has real users, the cognitive load comes back with interest. the speed is real, but so is the debt.
You move faster at the start, but you also accumulate ambiguity that you eventually have to pay down when the system gets more complex.
It feels less like coding and more like managing a probabilistic collaborator the cognitive load just shifts from writing to verification.