Once, I asked AI to explain a piece of code I was stuck on.

The answer came back so smoothly. It had examples, step-by-step logic, and a conclusion that sounded reasonable enough for me to almost nod along immediately. In a few minutes, the stuck feeling disappeared. I felt lighter, as if someone had cleaned up a messy room inside my head.

But about ten minutes later, when a friend asked, "So why does this part need a stopping condition?", I went silent.

Not because the AI's answer was wrong. It was not. The problem was that I had moved through the answer too quickly. I read it, it felt familiar, it seemed understandable, and I mistook that fluency for real understanding.

That moment was a little embarrassing, but necessary. It reminded me of something very simple: AI can bring me closer to an answer faster, but it cannot understand on my behalf.

The trap of an answer that is too neat

Before, when I learned a new concept, I usually had to sit for a while with confusion. Read the documentation, underline things, try examples, get it wrong a few times, and only then begin to see the path.

Now, with one prompt, AI can summarize something for me in seconds.

That is useful. I do not want to romanticize difficult learning so much that every form of support becomes suspicious. There are times when AI genuinely helps me get past a bottleneck: explaining a term with a clearer example, suggesting how to break down an assignment, or pointing out a syntax error my eyes had stared at too long to see.

But that convenience has a small trap: I can easily confuse "I read the answer" with "I built the understanding myself."

For example, in a class assignment, AI can help me outline a presentation. But if I only take that outline, change a few words, and submit it, I may survive a deadline while my ability to form an argument, choose examples, and defend my point stays exactly where it was.

At work, AI can help write a very polite email. But if I do not understand what I want to say, where the boundary should be, or what I am promising the recipient, that email is only a beautiful shell.

In code, AI can generate a function that runs. But if I do not reread every condition, test bad inputs, or understand why that approach was chosen over another, I am borrowing a result without building the ability to be responsible for it.

Real understanding is usually less polished than the final answer. It includes hesitation, mistakes, and going back to the beginning. But those ugly parts are what give my thinking roots.

Do not ask AI so I can avoid thinking

I have used AI in a slightly lazy way before, even if I did not want to admit it at the time.

When the assignment was hard, I asked, "Solve this for me." When a paragraph felt clumsy, I asked, "Make this smoother." When code broke, I pasted the whole error and waited for a command to fix it.

AI still helped. But in the end, I did not feel much better at the work. I only became faster at getting an answer so I could move on.

Later, I tried changing the way I asked.

Before asking AI, I write down my own thinking first, even if it is messy:

  • What do I think the problem is?
  • What have I already tried?
  • Where am I unsure?
  • If I had to guess, where do I think the cause is?

Then I ask: "This is how I am thinking. Please point out what is not solid." Or: "Do not give me the answer right away. Ask me a few questions so I can find the direction myself."

This is a little slower. But it keeps me awake.

AI stops being someone doing the homework for me. It becomes more like a patient study partner, willing to ask again, gently challenge me, offer a different example, or show where I am skipping a step in my reasoning.

And I realized something: the quality of an AI answer often depends heavily on the quality of my first thought. If I throw in a vague request, I often get back a vague-sounding answer. If I take the time to say what I understand, where I am stuck, and what I need to check, the conversation becomes much more useful.

Keeping my part of the work

I do not think young people need to avoid AI to prove they are hardworking. That is not realistic, and it is not necessary. Good tools should be used.

But I do think there are a few parts of the work I should keep for myself.

The first is curiosity. Before asking "what is the answer?", I want to practice asking "why does it work that way?" Why does this code fail? Why is this way of presenting more persuasive than another? Why does an answer sound good but still feel as if something is missing?

The second is judgment. AI can speak very confidently, even when the answer does not fit my context. So I still need to check: does this match the assignment requirements, does it fit real users, does it make the code harder to maintain, does it make me overpromise at work?

The third is personal effort. Some knowledge only sinks in when I do the work myself. Writing a bad first draft. Debugging a small error. Explaining the idea again to a teammate. Sitting with "I do not understand" a little longer before sending the prompt.

One small pattern I am trying is using AI in three steps:

  1. Make my own version first, even if it is clumsy.
  2. Ask AI for feedback, questions, or a different angle.
  3. Revise it myself and explain why I chose the final version.

The third step matters most. If I skip it, I am just handing the decision to the tool.

In studying, that may mean closing the screen and explaining the concept in my own words. At work, it may mean rereading the email AI wrote and asking, "Is this really what I mean?" In coding, it may mean running extra tests, changing inputs, or deleting one line at a time to see whether I understand what it does.

It does not sound glamorous. But real learning is often held together by small things like that.

AI makes me faster, and I am grateful for that. Some days it makes a blank page less scary, a stubborn error less lonely, and a new topic easier to begin.

But I do not want to move so fast that I leave myself behind.

I still want to keep the ability to wonder, doubt, try again, explain things myself, and take responsibility for what I make. Because if one day the tool changes name, the interface changes, and a new model appears, what remains with me will not be the answers I copied.

What remains is the way I think.

So maybe the question is not: "Should I use AI to learn?"

The gentler but harder question is: "After each time I use AI, do I understand a little more, or did I only move a little faster?"