A common worry: “AI will replace too much of your own thinking”. I think this risk is real, and acute.

I love to delegate: there are many things I absolutely do not want to think about. When coding: I architect, and the AI builds. I’ll have 3-5 agent sessions in progress. I rarely wait: there’s nearly always an output to review, or another plan to write. It’s fast, intense.

But let’s say someone asks for comments on a Google Doc. Often, the first step should be: paste the link into Claude and prompt “Thoughts? /ask-many-models″.¹ But: the wrong next step is to tab away and do something else while the agents work. Instead, I will:

1. Read the doc.
2. Voice note my own thoughts.
3. Read the AI outputs.
4. Voice note again.
5. Share both voice notes with the AI.
6. Read its reply, and continue the conversation.

In thought partnership, both partners should be thinking. And their first thoughts should be independent.