The point of the article as I see it is that incremental improvements lead to sudden changes,
when equivalence is reached in a specific domain.
It's a bit like rising water in a lake or river. You're fine, until you aren't, though the specific "aren't" moment will depend on your elevation relative to the body of water, of the height of any protections (levees, dams, storm walls), and any mitigating mechanisms (e.g., flood control pumps).
Up until the point that your defences are overtopped, your feet are dry. Once overtopped, you're wet.
Same with AGI (again, presuming that the present LLM/GD approach continues to provide returns): a slow but gradual increase in capability can subsume more and more human intellectual / cognitive tasks, and for each individual task, that moment is fairly likely to be rather sudden.