"Banning end-to-end encryption" implies banning/backdooring all serious encryption, reviving the long lines of failed attempts but in the context where serious encryption has wide, wide, wide adoption.
Earlier plan have quickly shown to be entirely impractical/counter-productive/corrosive. If you somehow enforced that, you can count on someone else besides the NSA getting the key. You would have to agree among all the world's states who gets the key or just create a "great US firewall" plus discarding all the infrastructure large companies have spent billions creating. You'd create a situation where "only outlaws have encryption", etc.
What other plans can we come up with? Maybe put software engineers in camps along with the refugees?
So much bad prose on the internet, it seems feasible to make this hard to distinguish from human-written text. Plus you could have multiple codings for values, allowing non-deterministic encoding but deterministic decoding.
How do you prove that the giant integer which represents this text is a random result of my thoughts rather than encoded information?
The second major side effect will be essentially making writing software into a completely corporate thing. Either the official algorithm is a secret (Remember the Clipper Chip and/or Skipjack?), or writing implementations becomes work that only a very trusted few are allowed to do (see inevitability of snooping above). Currently it's very difficult for medium to small companies to do the paperwork for security clearances. It's not going to be any easier to get programmers certified or bonded or whatever to work on official algorithm implementations. This means slower introduction of any innovations, and any buggy implementations will be in use for a long time before getting fixed.