Large language models have made it dramatically easier to write well. Ideas that once required hours of drafting can now be expressed clearly and elegantly in minutes. This is an extraordinary gain. But it also changes something subtle—and important—about the relationship between writing and thinking.
Writing has never been just a way of communicating finished ideas. It is one of the ways ideas become finished in the first place. The slowness of traditional writing plays a cognitive role. Writing forces intuitions to pass through resistance. In the effort to put a thought into words, we often discover that it is vague, overstated, or not quite what we meant. Many ideas do not survive this process—and that is precisely the point.
AI alters this dynamic by removing much of that friction. With the help of an assistant, even a partially formed intuition can be rendered into fluent, persuasive prose almost immediately. The resulting text can sound more settled than the underlying idea actually is. Expression becomes detached from filtration.
When ideas are articulated too quickly, they may not yet have been fully metabolized. They can be defensible, coherent, even elegant—yet still premature. The danger is not that such texts are “inauthentic” or “machine-written,” but that they create a false sense of completion. The idea looks finished before it has had time to compete with alternatives, attract objections, or mature through reflection.
My concern here is not how texts are judged by others, but how ideas come to belong to us in the first place.
AI also compresses the feedback loop of thinking. In traditional writing, the cycle of drafting, rereading, and revising unfolds over days or weeks. That delay allows doubts to surface and counterarguments to emerge. With AI, the loop can collapse into minutes. What gets amplified is not necessarily what is deepest or truest, but what is most verbally fluent.
Seen this way, AI is a powerful amplifier of expression, not a guarantee of intellectual solidity.
This leads to a simple view about how these tools are best used. AI is excellent at many things—brainstorming, stress-testing arguments, generating examples, even helping me see what I think by forcing me to react to alternatives. My caution is more specific: it begins when the tool stops assisting the thinking and starts producing the prose. AI is extraordinarily effective as an editor, but the first draft—the messy, slow, resistant one—is better written by oneself, precisely because that is where thinking still happens.
Only after an idea has been wrestled into existence does it benefit fully from refinement. Used this way, AI does not replace thinking; it helps complete it.
None of this contradicts the legitimacy of using AI to write better -see my previous post on this issue. On the contrary, these tools lower the cost of expression and allow more ideas to enter the public sphere—an unequivocal gain. The challenge is not to avoid AI, nor to fetishize a “pure” writing process, but to be attentive to sequencing. When generation outruns cognition, fluency can masquerade as depth.
The solution is not to slow AI down, but to slow ourselves back to a human pace before we decide that an idea is ready to stand on its own.
***If this essay resonated with you, you can subscribe to my Substack for regular reflections on economics, politics, geopolitics, AI, and literature.
*** Disclaimer: I used ChatGPT-5.2. as an editorial and language-refinement tool. The ideas and arguments are entirely my own, and I take full responsibility for them.