It doesn't work that way. The underlying assumptions change. The program can be perfect but still not matching what becomes necessary in a few years. The example they write about is exactly such a kind. Sometime in 2007 it was apparently slow to access files from the distributed storage on Google. Therefore, a Python script was written to fetch these files to some local disks. Python was obviously enough since that process was slow. Then, another program, written in C++ was to deliver the files from the local disks to the clients. The C++ program was then "sub-optimally" maintained, simply because it wasn't on the critical path for the company. There weren't even the good test cases to verify correctness of the changes.
Still, until 2013 a lot of assumptions based on which the program was written changed completely: Accessing files from the distributed storage became faster. The typical computer which is to deliver the data seems to have much more free RAM, allowing using RAM for caching purposes instead of the local hard disk. The nature of the load changed so much that accessing the local hard disk became the bottleneck anyway. Which is not so surprising: one disk seek is still around 10 ms, which means as soon as you can't deliver something from RAM you can deliver only up to 100 items from the disk per second, when they are not in the same block of data.
Don't tell me you have to write the program for "all possible assumptions." Unless you have unlimited time, you write it to solve the problem under existing assumptions. Once they change, you change the implementation.
Moreover, to achieve the changes related to the underlying assumptions, Go wasn't needed in the current example, but it was the most convenient approach for Brad, who is by the way obviously an extraordinary programmer.
What I agree with you is that if the program is bad even under the initial assumptions, it's certainly something that should be observed and tracked by the good manager: he must know that he can't rely on such program and that whenever the program gets any heavier load, the problems will happen.