He's extrapolating current progress into full mastery far too uncritically. AI is famous for having "long tail" issues.

He has a startup in this space, so his "expert opinion" is inherently biased towards "this thing I'm building is the bee's knees!"

Alignment for these systems is currently an unsolved problem. I'm not sure how he thinks this is a great idea.

His timeframe is delusional. It reeks of thinking from inside a SV startup bubble where you can Change The World(of California) in 3 years. There are huge swathes of the tech industry that don't exist in this bubble and move much more slowly.

That does not make it true or likely.

As a counter point, since he believes code review is not going away, how is one going to become a great code reviewer?

My answer is by learning to be a good programmer and asking good questions. Code review is a discussion between two or more people with awareness of context, goals and reasons behind the code as is: it's a collaboration.

But reviewing unknown code by unknown author requires such a tremendeous amount of work (and notably focus and concentration): that is so much harder without an ability to ask questions about reasons for each specific area of the code.