"Dark code" is behavior in production that nobody can explain end-to-end. Not just unreviewed code, but systems whose behavior emerges from interactions between components—often at runtime—without anyone ever holding a complete mental model of what the system actually does
So.. basically.. ai generated code that nobody understood ever.
Btw one of my thoughts was last year that we should indicate if some code was made by ai. But since then, depending on the project, sometimes all [or huge part of] the code is generated by ai.
Still, I think, it should be marked. Maybe not even with a binary true/false, but some estimation of level of human understanding. Similarly to code coverage.
De amúgy ha olyan a kódbázis, hogy emberek is írnak bele, akkor még érdemes lehet külön (pl. "Olivér AI asszisztense" ) perszónaként commitálni és akkor a blame-ből azonnal látszik, hogy mi AI és mi nem, és nem kell hozzá semmi special tooling. Ez persze akkor nem működik kényelmesen, ha egyazon commiton belül generálsz is kódot meg írsz is kézzel.