I might get worried once I see AI that's capable of maintaining existing software, maybe.
For now, AI can give you an implementation, but it can't tell you if that implementation is good, if it's secure, if it's fast, or if it will play nice within the ecosystem of an existing project.
We've got plenty of work over the years fixing code that was written by people who shouldn't have been trusted to write it, and we're going to get plenty more work fixing code that was written by AI that shouldn't have been trusted either.
