The point you made about agentic development mostly involving reading code rather than writing it really resonates with me. Supervising an LLM simply doesn’t lead to the flow state that results from building your mental model for a solution and then writing it yourself. It’s just not fun as a craft.