They argue you would be better with hardcoding your system to known secure best practice. When the time comes to change it, you specify a new protocol version, as there will be new and better practices not only in algorithms but also in how they are used (mac-before-encryption being the canonical example of this which took far too long to change).
This is a uniquely severe problem for hashing. MD5 is by far the fastest of the common hashes, at least 20% faster than SHA1 for example.
The Venn diagram is you've got popular fast hashes, of which md5 is arguably the best, and in fast hashes, you've got cryptographically secure fast hashes, of which md5 most certainly is no longer a member. Its awesome for everything non-crypto non-secure non-over the internet.
Last weekend I had to burn a legacy DVD and I compared the md5sum of the image to the md5sum of the burned DVD, thankfully they matched. As much as I distrust legacy media, there probably isn't a sentient opponent in the burner, although I've occasionally sworn otherwise, and the most likely failure mode would have been simple truncation or buffer dropouts, so sheer speed was the priority. The hashes obviously matched, it was a good burn.
I've also used them as checksums for engineering data files. Everything on the planet from AS400s to MySQL databases can calculate md5, which is convenient for cross platform interchange type stuff. Heres a data blob and its md5, does your hash calculation match? Oh it does, how nice to know you have perfect data integrity, here have the next one. On the internet I wouldn't trust the opfor with md5, but I can trust my own coworkers and my own engineering machinery not to try and attack me (well, probably).
That means until the end of time you'll unfortunately have people discover md5 is available for non-crypto use, then try to use it to hash passwords or DRM executables or something, which is not so wise. Or for a downgrade attack, they misdesign their crypto system to allow any hash installed on the machine as the hash type and not blacklist md5 and sha1 and maybe more.
What will die out is hashes like SHA1. Why use something almost as dangerous as md5 for crypto, thats slower than md5, and not as widespread as md5? Bye bye sha1!
