A sufficiently detailed spec is code
haskellforall.comWell done for actually trying and failing to reproduce their results.
But the point from the original comic version kind of makes it seem like specs are useless, which I don't believe is true. Just like reimplementations can be useful (directly so for property testing or golden testing, but also for learning what things worked and what didn't), a specification can be useful even though it is informal or incomplete in that it guides the real implementation. (Like "plan mode", for you vibers.)
It's so easy to make a specification which can't be implemented (or shouldn't be). Very often, I go back and tweak the specification, after I start the implementation and figure out which assumptions are wrong ...