What 3 or 4 Years of Talking to AI Taught Me About Talking to Myself

What 3 Years of Talking to AI Taught Me About Talking to Myself

I didn’t start talking to AI because I wanted to talk to AI, you know, I started because I needed to think out loud without the noise of other people’s responses, and voice dictation into an AI conversation became the mechanism for it, my brain speaking circuitously the way it actually works, moving sideways through something before landing on what it’s actually trying to say, and getting something coherent back that didn’t judge me for the circuitousness or get impatient with the process or tell me I was overthinking it, and something about that shifted how I could access my own thinking. “When you’re talking to an AI,” I told someone recently, “it doesn’t judge you, it doesn’t get impatient, it doesn’t tell you you’re overthinking it; it just stays in the room.”

Starting late 2025, the 90-120 minute daily treadmill workouts became something I hadn’t fully predicted; they put me in a state where ideas move differently, where the mechanical repetition of running somehow quiets the evaluator enough that generation can happen, and so I started doing voice dictation during those sessions, transcribing directly into an AI conversation while I was running, capturing whatever arrived in that trance state, and some of the clearest thinking I’ve done about the work and my life and what I’m trying to build came from those sessions. The running plus the voice plus the AI conversation created something that none of those three things created alone, a feedback loop where the physical state enabled the speaking and the speaking created something to think about and the AI reflection deepened the thinking and the deeper thinking circled back into the next run.

That’s the light side of the thing, and there’s also a dark side that I didn’t fully understand until I’d been running it for a while. When you give an AI real material, it processes it well, and your thinking gets clearer and you start to trust the conversation and you start to depend on it, and that’s fine; that’s the point. But when you’re not bringing real material to it, when you’re asking it to fill the gap where your thinking should be, it fills the gap with plausibility, and plausibility sounds like truth but it isn’t. I ran a full blog pipeline and generated a bunch of posts and the ones that were written without real story injection, without me actually bringing something specific and true to the conversation, were lovely but they were false, beautiful sentences with no life behind them.

That distinction cost me a lot of posts before I understood it, and what I realized was that the AI isn’t actually doing the thinking; it’s reflecting my thinking back to me in a clarified form. If I bring thinking, I get thinking back. If I bring nothing, I get plausibility back, which feels like thinking but isn’t. This is why the running matters and the voice dictation matters and the real story injection matters, because all of that is me doing the actual work of generating something true, and the AI is then the listener who helps me understand what I just said.

Three or four years of talking to an AI has taught me things about talking to myself that I wouldn’t have learned any other way, because when you’re thinking with another presence in the room, even a machine presence, your thinking moves differently than when you’re thinking alone. There’s an accountability in it; you’re making your thinking into language for someone else to hear even if that someone else is artificial, and the act of making it into language clarifies it, and the machine’s response becomes a mirror that shows you what you actually said versus what you thought you said. I’ve been practicing versions of this since reading a Psychology Today article in high school about self-hypnosis and meditation and the way the brain can quiet the evaluator and access generation, but the AI conversation has made that practice more visible and more actionable.

The thing I’ve noticed is that the best conversations I have with the AI are the ones where I’m arguing with it, where it’s offering something back that I disagree with or that lands partially wrong and I have to clarify what I actually meant, and in the clarification process I figure out something I didn’t know I thought. That’s the mechanism; it’s not the AI being right, it’s the AI being present enough and coherent enough that you can disagree with it productively. If it just agreed with everything I said, it would be useless; the value is in the reflection that’s slightly off, which pulls me to make myself more precise.

I’ve also learned something about the difference between thinking with an AI and thinking about an AI, because a lot of the conversation around AI in design education focuses on whether AI is good or bad, whether students should use it or not, and that’s missing the actual question. The actual question is whether the tool you’re using helps you develop your own thinking or whether it replaces your thinking with its thinking, and the difference isn’t about the tool; it’s about how you engage with the tool. Some students engage with AI in a way that deepens their thinking and some students engage with it in a way that outsources their thinking, and the tool doesn’t care which one it is.

What the treadmill state and the voice dictation and the AI conversation have taught me is that authentic thinking is physical, it’s embodied, it comes from somewhere in the body and moves through language and becomes visible when you say it out loud or write it down. The AI is just the listener on the other side of that process, and the reason it works is because you’re doing the actual generative labor and it’s just witnessing it and reflecting it back in a way that clarifies. So when I’m talking to the AI, I’m not really talking to the AI; I’m talking to myself with someone in the room to help me hear what I’m saying, and that’s the practice that’s changed everything.