Every product team knows the feeling: a spacing issue that would take 30 seconds to fix in a design tool takes half an afternoon once it’s in production code. Review, CI, deploy, verify. The work didn’t get harder — the medium did.
Prompt-to-code tools accidentally made this worse. When generation goes straight to code, every iteration — every “make the header smaller” — happens in the expensive medium. You’ve deleted the cheap phase of the funnel and kept the costly one.
Run the numbers on a typical flow. A design-phase iteration costs pennies of compute and seconds of attention. The same change post-generation costs a regeneration (with its bloat and drift) or an engineering round-trip. Across the dozens of iterations a real interface needs, that’s the 50x gap we keep citing.
The fix isn’t fewer iterations — iteration is where quality comes from. The fix is keeping iterations in the cheap medium for as long as possible, and paying the conversion cost to code exactly once, when the design is settled.
That ordering — design first, code second — is the entire premise of Nokuva, but the analysis holds whatever tools you use: measure where your iterations happen, and move them to the cheapest medium that preserves fidelity.
