Recently, I caught myself doing something that has become almost automatic: I had a question, opened an AI conversation, and got a clear answer before I had finished working out what I thought.
It helped. But it also left me uneasy: if I always reach an answer this quickly, when do I do the work of understanding it?
The question is not whether using AI makes us less intelligent. Put that way, it sounds too much like a headline designed to scare us. The question I care about is more specific: which part of our mental work are we no longer practicing when the tool makes the first move for us?
A finished answer is not the same as having thought it through
Two people can deliver nearly identical reports. One compared sources, challenged hypotheses, and changed their mind after finding an inconvenient fact. The other copied a convincing answer and polished its style. From the outside, we see two documents. Inside, two very different processes took place.
Critical thinking is not distrusting everything or disagreeing for sport. It means asking where a claim came from, what evidence supports it, what alternative explanation exists, and what would make us change our minds.
AI can participate in that process. It can also let us skip it. Its greatest strength—removing friction—is what makes the difference hard to notice.
A well-written explanation creates a feeling of understanding. It happens to me, too. I read something, nod, and feel that the subject is settled. Then I close the conversation, try to explain it to someone else, and discover the gaps. What I had was a good answer in front of me, not necessarily knowledge of my own.
What we know, and what we do not
Research does not support a universal claim that “AI destroys critical thinking.” A review of 25 empirical studies in education found both patterns: structured activities that asked students to question and revise AI responses supported analysis, while less-guided use was associated with passive acceptance and lower cognitive engagement.
Another review of STEM education included 85 studies, 49 of which qualified for meta-analysis. The average positive effect on learning initially looked promising, but results varied enormously and the overall benefit weakened after accounting for possible publication bias. The authors found a more useful distinction: whether AI expanded or replaced the student's cognitive activity mattered, as did whether the outcome measured was knowledge or skill.
Ten of the reviewed studies documented students accepting incorrect answers without questioning them. That does not show that AI has “switched off our brains.” It does point to a practical problem: checking an answer requires subject knowledge and a verification strategy. Simply saying “verify everything” is not enough, especially for someone who is still learning.
These studies mostly examine educational settings. They do not prove that the same effects occur for every professional, task, or period of life. But they help us ask a question worth bringing to our own desks: is this assistance building my capacity, or just delivering a result faster?
Delegating friction or delegating judgment
I do not feel that I lose an essential skill when AI organizes my notes, fixes a comma, or reminds me of syntax I already understand. Some repetitive work is worth getting out of the way.
The situation changes when I hand over the very part I wanted to learn.
If I am researching why a company failed, I can request a summary and accept its conclusion. Or I can write down three hypotheses first, ask AI to find objections to each, and then examine the sources to decide which explanation holds up. I used AI in both cases. Only in one did I keep responsibility for the reasoning.
Programming offers a similar example. Asking for a draft of code can save time. But if I am trying to learn a concept, I should be able to explain why the code works, test difficult cases, and recognize where it might fail. Otherwise, I may have obtained a solution without developing the ability to produce or evaluate the next one.
I am not suggesting we do everything by hand. Giving up a useful tool just to prove we can still suffer would be absurd. I am suggesting that we distinguish between solving a task and developing a skill. Sometimes those are different goals.
A demanding conversation partner, not a compliant substitute
One way to use AI better is to make sure it has neither the first nor the final word.
Before asking about something important, I write even just two lines: what I believe, what I do not know, and what evidence would change my mind. That gives me a point of comparison. If the model answers first, it becomes too easy to adopt its assumptions without noticing.
Then I can ask for something more valuable than a conclusion ready to copy: “Find the weak point in my argument,” “show me the strongest objection,” “ask questions that reveal whether I really understand,” or “which primary source should I check?” AI starts to look less like a task-finishing machine and more like a conversation partner that forces me to sharpen my ideas.
Finally, I try closing the conversation and making something on my own: explaining the concept, making a decision, writing a paragraph, or solving part of the problem. Not as punishment. As a test of whether anything from that exchange stayed with me.
Verification should also match the risk. A movie recommendation does not need an investigation. A medical or financial claim, or a decision that affects other people, deserves stronger sources and, where appropriate, professional judgment.
It is not about using AI more or less
We can spend hours with AI comparing arguments and end up thinking more than before. We can also open it once a day to avoid making any attempt of our own. Counting minutes of use tells us very little.
The question is what we delegate. If we hand off repetitive work to devote more attention to judgment, the tool extends our abilities. If we hand off judgment to produce faster, we may increase our output while losing some autonomy over how we evaluate it.
We do not need to turn every AI query into a debate. Some days we just need to get something done and move on. But we should recognize the occasions when an immediate answer would take away the practice we most need.
So, is AI making us think less? Sometimes, yes. Especially when we let its first answer do the work of our first question, too. But that is not an inevitable property of the technology. We can use it to find better objections, clarify our doubts, and practice until we understand.
Perhaps the most important habit is to pause before writing the prompt and ask ourselves: do I need an answer today, or do I need to learn how to think through this problem?
In this video, I explore the question further and share more examples of using AI without handing over our judgment.