Except it is true in the sense that level of control teaches you where the bottlenecks are, and without that "feel" you increasingly lose insights that don't cascade down when someone only ever trains on/uses tools that abstract away the nuance that you get further from what the work is about
this is a good post and I don’t doubt that LLMs pose a problem for the author‘s preferred style of working, but I wonder if there’s a bit of self-deception in the conception of always wanting to work on the lowest level thing possible. this can’t really be true in practice