Notable examples are the user asking the AI to explain what an obfuscated function does in human terms, and then being asked to rewrite it in more readable terms.

This task seems pretty useful. It's yet to be seen how well it can do the same for entire codebases. Handling one function is achievable for humans, albeit much more slowly. Needless to say, I figure it's more than capable to do so with little extra effort on the input side; it may already be capable of doing so as is.

Another interesting thing was an example where someone had it write a C syntax highlighter. This is where things get interesting. Syntax highlighters have been written before. The AI isn't doing anything here but literally copying a C syntax highlighter from its training set. Yes, I know it's possible it wasn't given a syntax highlighter in its training set, but it was given many things that are at least very similar.

That's the key: the AI is copying code from its training set.

That's how AI works, at least for now. When it is doing anything, it is copying and modifying from examples. The interesting thing about code is that it is uniquely suited for this type of copy-paste-modify activity. It's how many programmers already work.

If this continues as expected, I anticipate we'll see a tool that helps generate smarter boilerplate in projects. Imagine having an AI code generator helping you out in your IDE. If or when things progress beyond that, then sure, maybe coding jobs would be in peril. 

But that may be true for different reasons.