I have been using AI a great deal lately. Every day brings a large number of conversations: writing code, handling work, organizing my blog, and discussing questions I have not fully worked out yet. When I put it to work on something concrete, it is usually useful. It helps me get certain things done faster.
But the more I use it, the more I notice that different tasks cannot all be handed to AI in the same way. When I organize a blog post, the facts, feelings, and opinions are mostly mine; AI helps me express them more clearly. But if I try to use AI to learn something I know nothing about, the situation is much less simple.
The process itself works well. Something I already know, but have not yet organized, can become clearer after a few rounds of conversation. A relationship I could only sense may become a complete argument. A messy piece of writing may find a structure that works. AI is a little like a patient editor, helping me look at what I already have one more time.
But its additions can also feel strange.
For example, I might say only that I have been doing many things lately without knowing which ones really matter. AI can quickly turn that into a problem of “scattered attention,” “the illusion of efficiency,” or “productivity conversion,” and add a complete analysis. None of the sentences looks obviously wrong. Some sound more professional than anything I would have written. Yet after reading them, I find myself asking: is this actually the problem I am having? How did I get from what I just said to this conclusion?
The article has become more complete, but I may not understand myself any better.
Where the idea comes from
The most important part of writing may not be structure, but why I wanted to write about something in the first place.
An idea is not just its final conclusion. It also includes what happened to me, what facts I saw, why I had a particular reaction, and where I think the problem actually lies. These things can be scattered, sometimes no more than a few sentences that do not look like an article yet. But they are the reason for the article.
AI can help me organize this material. It can notice a connection between two passages, point out a counterexample, or explain what I mean in another way. That kind of addition is valuable. Sometimes I need another perspective in order to see my own thought again.
But the addition is not transparent. From the language I provide, AI will automatically look for concepts, complete causal links, and arrange a hierarchy. A moment of hesitation can become a general problem. A feeling that only makes sense in one specific situation can become a broad principle that sounds convincing.
Those sentences may not be wrong. They may simply not belong to me. An idea assembled from many possible expressions can certainly be valid, but it does not necessarily have my experience and judgment behind it.
The feeling of being pushed along
I used to assume that a more complete article meant more progress. Now I am increasingly suspicious of that feeling.
Talking with AI makes it easy to remain in a state of “doing something.” A moment ago there were only a few thoughts; a moment later there is a title, an outline, paragraphs, and a conclusion. I can then ask it to polish, expand, adjust the tone, and generate several more versions. Each action is easy, and each one looks like progress.
But AI can easily drift, and the drift may not be obvious right away. It can pull a specific question into a larger concept, or keep extending a direction that merely looks reasonable. This does not happen every time. It is more like a probabilistic result: once I enter that kind of workflow, a small deviation can be amplified much faster.
I was already used to working across several threads. In a day I might be talking with nearly ten people while moving five or six other things forward. AI amplifies that state. Every task gets an immediate response, every direction can keep unfolding, and something that only occasionally went off course can travel much farther through continuous generation and continuous replies. I remain in a work state the whole time.
But greater intensity does not mean that more important things have been done. Sometimes only the text is moving forward, while I am no closer to the question. I am being pushed from one version to the next, from one explanation to the next, until I have an article but cannot say what I actually understand better than before.
This is what I mean by the illusion of busyness. The content increases, but my understanding does not increase with it. Being able to deliver a result does not mean that I understand the result. Being able to follow an explanation does not mean that I could make the same judgment later on my own.
Some people call content that looks like an outcome but does not move a task forward “AI slop” or “workslop.” In personal use, there is a quieter version of the same problem: the content may not be bad at all. It may even have been polished enough to pass. But I have not really formed a relationship with it. I have only kept confirming, revising, and accepting what AI gives me, until I feel that I have been working all along.
I can borrow the structure, not the thinking
AI is good at organizing an article. It may arrange the material as background, problem, analysis, and conclusion. That is often clearer than my first draft, and it can make the piece easier to read.
But I do not want the structure to decide my thinking in return.
Human thought does not always begin with a polished concept. Often it begins with a specific person or event, followed by a feeling I cannot quite explain. Only later do I find the question and form a judgment. If I ask AI to turn everything into a standard argument at the beginning, the article may become smoother while losing the path that led me to the idea.
So I can borrow AI’s structure without handing it my thinking. It can help me notice relationships, but it cannot decide which relationships matter. It can add a perspective, but a perspective does not become my judgment simply because it is well phrased.
What I approve of is not always what I mean
Sometimes an angle suggested by AI is the first thing that makes me notice a possible idea. That does not mean it can never become mine. People discover new thoughts through conversation all the time.
The difference is whether I understand it again after discovering it. Can I explain it without AI’s original wording? Can I say what fact it came from, what problem it addresses, and where it applies? If all I can say is “that sounds right,” without being able to say why, it may not have become my thought yet.
I now keep checking what AI has added. Is it helping me say what I already have more clearly, or has it carried the article somewhere else? Is it offering another perspective, or deciding what is important on my behalf? Am I saying “this argument holds,” or “this is what I want to say”? The two sound close, but they are not the same.
AI can make what I know clearer
I do not want to return to a life without AI. It has become a very useful part of how I write.
I am simply clearer about where its value lies. It is not mainly a way to write things I have never understood. It helps me see more clearly what I already know. My experience and facts are the material; my judgment gives the direction. AI helps me find connections, look for counterexamples, and try different expressions. After several rounds of conversation, something I could previously only sense may become an idea I can explain, repeat, and keep discussing.
This is why I am uncertain about the phrase “using AI to assist learning.” AI can help me build a path for learning: break a question down, arrange a route, find exercises and feedback, and explain something in another way when I am stuck. But it cannot actually learn a field for me, and one conversation does not become my knowledge just because I understood it at the time.
More precisely, AI is also a training process for the person using it. It is not only answering my questions; it is training me in what I will do the next time I face one. If I ask for the answer every time, I may be training myself to skip accumulation and judgment in pursuit of a complete-looking result. If I ask it to question me, make me explain the idea again, and test it with new examples, I may instead be training understanding and judgment.
Knowledge still has to accumulate. Being told a concept does not mean it has entered my system of knowledge. Understanding an explanation does not mean I possess the judgment inside it. Real understanding still needs detail, process, counterexamples, practice, and feedback from the real world.
If I have no experience with a subject and simply ask AI to generate a complete account, I usually get an overview. It can tell me how people generally talk about the subject, but it does not mean I understand why they talk that way. It will not automatically make my observation sharper or my judgment clearer.
That is why I try to say my own thing first and then let AI help organize it. After writing, I also step away from the conversation and see whether I can explain it in my own order. If I can only follow the sentences it gave me, I have a polished result. If I can explain the facts, logic, boundaries, and judgment, the article is genuinely mine.
In the end, I still have to return to reality. Whether these words helped someone understand something, make a decision, or cause something to happen matters more than whether the prose is complete. I increasingly think this is the real center of the issue: attention and judgment matter to people, and they matter to people using AI as well. AI can rapidly expand the direction I choose, but it cannot decide whether that direction is worth continuing.
AI can help me produce content faster, and it can help me see thoughts I already had more clearly. But it does not automatically bring understanding, and it certainly does not automatically bring progress. It can supplement my expression, but it cannot confirm that the expression is really mine.
I still do not have a complete answer. I am simply becoming more alert to the feeling of being carried along by content. The article is finished, and it seems that the task is finished too. But I still want to ask:
AI writes well, but is this really what I mean?