Aye, so this is fine as long as there's just one entity generating IDs, so it can automatically check for collision and re-generate.

In a distributed system it's another matter.

I'm currently using 6 base-40 chars (upcase and lowcase letters, so e.g. "SXJwzQY") that represent an integer Unix time.

The base-40 alphabet is enough that all IDs up until year ~2099 will fit in 6 chars, and it's nicer to type on the phone when you only have to switch between two keyboard modes, the upcase and lowcase but not also the numbers and symbols mode.