AI can do some amazing things inferring from past behavior, but it certainly is a fair distance away from anything approaching understanding what the end goal is or what all the constraints are or might be in the future. Sure, people can set up some boilerplate problems and get custom boilerplate solutions but I really see AI as another level of tooling/abstraction that will only empower engineers to work less on the easy stuff and spend more time at a higher level of concern.

In other words, I think the near term gain is that we will actually ACTUALLY not have to worry about centering a div anymore.