Seeing the words "recursive self-improvement" I was expecting something else from the article. E.g. how the transformer architecture or agent design is being changed/improved through LLM automation, but the article mostly talks about the LOC counts.