Most people use AI like a tool. Ask a question, get an answer, move on. That’s fine — but it’s like using a piano to play one note at a time. Here’s what thinking with AI actually looks like, based on what I’ve noticed works.
Throw ideas without filtering. Don’t polish your thoughts before sharing them — say the half-formed thing, say the dumb thing, say the thing that contradicts what you said two minutes ago. AI doesn’t judge you and doesn’t remember your reputation; it takes whatever you give it and works with it. The raw, unfiltered input is better input, because it’s closer to what you actually think before your internal editor kills it.
Interrupt yourself across domains. The best ideas come from collisions between unrelated topics, so don’t force yourself to stay on one thread — jump from email to philosophy to career strategy to medicine and trust that the connections will emerge. The key is a collaborator that can hold all the threads simultaneously: a human collaborator would lose track, a notebook captures but can’t connect, and a long-context AI does both.
Challenge the tool. The most important thing you can do with an AI is disagree with it. Ask “are you just agreeing with me?” Ask it to steelman the opposite. Push back when the answer comes too fast. AI is biased toward agreement, and if you don’t actively fight this you’re in an echo chamber that feels like insight. The collaboration only works if you maintain healthy skepticism about your collaborator.
Notice the patterns. Pay attention to what’s happening in the conversation, not just what’s being said. When three separate threads suddenly connect, that’s a signal. When you feel energised, that’s data. When the AI keeps agreeing, that’s a warning. This meta-awareness is what separates thinking with AI from just using it — you’re not delegating your thinking, you’re amplifying it, and that requires knowing what your thinking is actually doing.
Don’t over-plan. The gap between idea and action should be as small as possible: “write it” is better than “let me think about whether to write it.” You can always edit later, but you can’t recover the energy of the moment later. AI makes this easier because execution is cheap — a garden post takes 30 seconds of your time (you say “write it”) and 30 seconds of AI time, so the cost of trying is nearly zero while the cost of deliberating is losing the thread.
And stop when you’re tired. Your cognitive state determines the quality of the collaboration: when you’re in flow, ride it; when the energy drops, stop. Don’t push through — the output degrades and you won’t notice, because the AI keeps producing confident-looking text. The AI doesn’t get tired. You do. Knowing the difference is the skill.