All AI programs to this point belong to the subcategory of "narrow AI" and thus there will be certain issues that coders will be required for in order to fix as AIs are fundamentally lines of codes themselves and are very limited in their flexibility.

For such a system to work properly, the AI needs to be bug-free. Since most times this is not the case, the AI is vulnerable to producing broken and buggy problems. So some oversight is still necessary and will be necessary.

You could write an additional machine learner to uncover bugs that people can fix, but this remains a very hard problem in the space.

And as you mentioned, we'll still need coders to write new creative programs since AI mechanisms are unable to create "novel" programs even if they have the components to do so. Novel meaning a program that exists outside of its domain of programmability (as limited by its instruction set).

There also remains the problem of intentionally and semantics. AI aren't able to know what a program's purpose is while coders know this before they write a single line of code. Without knowing this intent, they could produce code that is less flexible than desired by clients of the code. Also with semantics, there could be issues with translating a clients desires into inputs the AI can easily understand.

Because of this, "good/creative" coders will be safe for a while (I would estimate 20+ years), as many coders will move from raw coding to quality assurance, idea development, and maintenance roles.

The people who are the most at risk are the non-coding non-creative types. Especially those that undertake menial tasks that require little abstract thought.