Your comment spawned the following (minor?) idea for me: A lot of the devices (and hence IP addresses) we basically know are never going to be accessed directly

Your comment spawned the following (minor?) idea for me: A lot of the devices (and hence IP addresses) we basically know are never going to be accessed directly by a human being, only through some combination of tools and automated processes, if at all. Make a distinction, where possible and appropriate, between these and the actually-UX-relevant addresses and reserve the shorter/nicer ones for human use. (This may not be a great idea, or may be impractical, but I still wanted to put it here, because there's benefit in mapping out any amount of the potential solution space.) Also, thinking about it in this way, "end user" for IPv6 does not only translate to the developers working with IPv6 as consumers (think someone getting an IP from AWS for deploying their app), but, for maximum effectiveness, should also include the developers working on the implementation as well as distribution end of the technology. As always, these two (or more) groups have different levels of motivation to adopt and use the technology, and as such it is likely that the usual tradeoffs can be made (making something a bit harder to implement, for the sake of much increased "consumer" experience). Ideally there would be a spec and an implementation that would appeal to both allocators as well as consumers of the address space.