Great post. I largely agree with all points, though I do think using an IDE with LLM integration has some QoL value as long as you don’t let the agent/LLM go nuts making changes. Do you provide any sort of guidance or constraints doc when you start a session with an LLM?