When Is 100% Vibe Coding OK ?

Viorel PETCU Posted on May 31           When Is 100% Vibe Coding OK ? # ai # go # cschallenge # productivity Recently, I lost a lot of time reviewing and would have lost even more implementing, what I can only describe as vibe-written specs. Requirements that felt coherent while saying almost nothing concrete. The kind of document where every paragraph sounds reasonable until you scratch the surface a bit and realize the thing is not applicable in the existing ecosystem. To be clear: I am not against vibe coding . I use it myself all the time. If I need: a one-off script, a throwaway visualization, a temporary parser, a document nobody else will maintain, then speed matters more than rigor. But maintenance changes the equation completely. This weekend accidentally gave me a good example of where vibe coding actually shines. During my usual Saturday clean-up, I knocked over one of those wooden cube puzzles: 25 identical pentacubes that are supposed to fill a perfect 5×5×5 cube. Specifically this one: I tried reconstructing it manually for a while and failed ( blaming the fact that I was tired and there was heat dome over EUROPE messing with the comfort index 🥵 ) Needless to say this really bugged the engineer in me and on Sunday morning I was ready to throw AI at this. I realized that: the rules are precise, the state space is finite, the problem is absolutely solvable, and computers are very good at brute force. So I opened Go and started vibe coding . The single piece shape is simple: A planar pentacube of which there are 25 instances. At first the challenge was not the algorithm. It was the representation or the domain specific language ( DSL ). Even explaining the puzzle to ChatGPT took multiple iterations. We had to establish: coordinate conventions, slice notation, legal piece states, rotation semantics, and eventually a labeling system for pieces. Once the model stabilized, the rest followed naturally: generate transformations, normal