What I've learned in 2 years of heavy LLM use - ChatGPT, Gemini, and Claude, is that the significance is on expressing and then refining goals and plans. The details are noise. The clear goals matter, and the plans are derived from those.

I regularly interrupt my tools to say, "Please document what you just said in ...". And I manage the document organization.

At any point I can start fresh with any AI tool and say, "read x, y, and z documents, and then let's discuss our plans". Although I find that with Gemini, despite saying, "let's discuss", it wants to go build stuff. The stop button is there for a reason.