I work in AI for a living. I believe in the tools, use them every day, and know how much they unlock. But lately I have been wondering about a side effect that feels easy to miss: we are starting to treat our own thinking like a slow API call.
In the last 6-12 months, I have been able to work on so many things that once felt almost out of reach. I always wanted to write more - and not just more, but better. But I was never a naturally great writer. I needed someone to brainstorm with, fix my copy, help with vocabulary, and think through the different directions a blog could take. With Claude/ChatGPT, that kind of help is available whenever I want, however I want, on my schedule. That is a superpower.
And yet, underneath all that productivity, I sometimes catch a small fear: what if I slowly lose the ability, and even the interest, to think something through on my own, just as many people already seem to be doing? What if I stop sitting with my ideas? What if I start typing into the chat window before the thought has even fully formed, just to find out quickly what my AI thinks? Honestly, sometimes that does happen. I will be out on a walk, have a half-shaped thought, and feel that quick reflex to ask AI right away. So far, I have been trying my best to resist that urge — to delay the inference.
We all have the tendency to naturally offload mental effort when a reliable external system is available, because working memory is limited and thinking is costly. That is normal. Search engines did the same thing for information retrieval: they helped us get answers faster. But there is a difference between using a tool after thinking and reaching for it at the very first spark of thought. Some of our best ideas do not come when we are actively producing. They come when the mind has room to wander, to connect memory, emotion, and pattern quietly in the background.
Researchers often associate that with the default mode network and its interaction with executive-control systems in creative thought. If every vague instinct gets handed to AI too early, maybe what disappears is not intelligence, but that internal wandering stage where original thought starts to form.
The phrase I keep coming back to is this: delay the inference. In AI systems, inference is the step where you move from input to output - the model takes your prompt and generates. In human cognition, inference is how you arrive at a conclusion. Both meanings matter here. The habit worth building feels similar in both cases: do not run the model yet. That is how I have started approaching my writing. For months, I have been collecting raw thoughts in Apple Notes for this blog. Then I carved out time to write a really rough version myself, using those notes. After that, I did some research to understand how this behavior might connect to brain science, and only then did I go to AI to brainstorm the premise. It gave me a few interesting angles. I took a day to sit with that feedback, and then I came back and wrote the blog again — this time directly in the chat window.
What I am trying to say is simple: when the impulse to prompt arrives, notice it first. That pull toward the chat window is not always a productivity instinct. Sometimes it is the same thing as checking your phone the moment a conversation gets quiet. The thought does not need to be validated yet. It needs to be yours first. The inference — in both senses — can wait.