This was actually one of the design goals of UTF-8. A decoder should be able to pick up a stream at any point and not lose a single valid codepoint.
The physical part of disks and networks usually have things like preambles, correlators, phase-locked loops, Gray codes, and self-clocking signals which can be used to deduce bit and byte boundaries, which is why in software, we almost always deal with information which we already know is correctly aligned.
You need to add extra bits so you can find the alignment, and once you do that you have the answer to your question.
The original paper also discusses how to use the "Eliminate all zero bytes" behavior in order to set up an unambiguous sync pattern to indicate the beginning of transmissions in a bit-level wire protocol.
He attempted to answer it, but the answer assumed that the recipient could recognize deleted characters, which doesn't make any sense to me (it seems to be rather like assuming that if someone sends you a postcard and you don't receive it, you nevertheless sense, at the time it would have arrived, that you should have gotten it).