The real-world consequence of this would be people not writing tests until after they are feature-complete so they don't risk losing work.
Anyway, I suspect he's proposing it not as a serious methodology, but as a thought experiment to spur you into how you can make your changes smaller, simpler, and more atomic.
> I don’t suggest you try “test && commit || revert” because it’s better than what you do now. I suggest you try it because:
> - It’s cheap.
> - You’re bound to learn something.
The few projects that fully embraced it worked surprisingly well.
People's preconceptions about what XP is are unworkable, sure, but that's on them.