I think, the problem is that the customer-provided spec is so far away from being precise, that it just isn’t useful.

You end up making tons of micro-decisions while implementing and then you go out into the real-world to test your assumptions.
At that point, some micro-decisions will be wrong, but the vast majority get confirmed by simply not causing problems. And your code ends up as the closest approximation to a precise specification, because it’s the only documentation of many implicit requirements.

Somebody please, present non-imperative languages to that blog author…