AI Is Making Me Too Lazy to Think and Too Reliant on It

The Problem

For the past few days, I have been adding features to my website. As I mentioned when I tore it all down and rebuilt it last time, I used AI to recreate most of what the old version had, moving its features and interface to a rewritten version built with a new stack. Up to 99% of the code was not written by me at all, but by AI. Still, I had to understand the core because I was the one guiding AI in that direction; I did not want to let it decide everything on its own because that could easily get out of control. Article data is stored in an SQLite file, then 11ty extracts it to generate completely static HTML pages. A script filters that SQLite data into a smaller file, uploads it to GitHub Pages, and uses it to fetch dynamic data. I covered that whole process in the previous article; you can read it again here: New Stack: 11ty, sql.js-httpvfs, and Preact.js.

While starting the project, an error appeared: places that relied on dynamic data did not show up, and the API did not work as expected. Out of habit, I opened Claude and typed: "Check why ... is not working." More than ten minutes later, Claude gave me a quick summary and suggested relevant actions. I agreed, and within a few minutes the work was done and the website was functioning normally. A few years ago, a small mistake like forgetting to add an environment variable to the .env file could have had me spending hours inspecting every place to find the cause. Now, everything has become much simpler: open Claude, ask, get an answer, and finally approve the action. It happens so naturally that it feels like a reflex, without needing to think too much.

Then came a day when the process was no longer that simple. All it took was an issue involving an AI provider, or just a few minutes with my hand on my forehead wondering: "Do I really understand all the logic in the new blog's codebase?" I opened a source file containing API logic. The blocks of code looked neat and tidy, but looking at them made me feel nothing. Who wrote them? What does this line mean? Why was it written that way? "But it is probably fine. If something breaks or a new feature is needed, AI will have to keep writing it, not me." That was how I reassured myself.

As someone inclined to overthink, I am always torn between the boundary of speed and process. Who does not like working faster, but does everyone also like being too lazy to think? The faster the pace, the more the process gets burned away, leaving us little time to look back at what happened. I have always wanted to find ways to improve my productivity, get more done, or at least clear the backlog quickly to make room for other work. When AI appeared, it felt like a miracle pill for the sluggishness I had been struggling with for so long. But why, after all that time, do I feel even more tired than before?

So what gets lost when speed defeats the learning process?

What Has Changed in How I Solve Problems

Before LLMs, I would usually solve problems by reading documentation and articles on the Internet, where many people had encountered similar issues. I played the role of the processor: identifying the right keywords, synthesizing multiple sources, trying, failing, and trying again until things worked and I arrived at the conclusion: "Oh, so this is how it has to be." What I tried beforehand was not entirely useless; it helped me understand why it did not fit, while the right solution felt like a reward after hours of sweating over it. Often, finding the answer was not even the final destination, because the knowledge gained during the search was what mattered: understanding which approach to use in which case. Knowledge is not a straight line; it keeps branching out and connecting closely with itself. There was a time when I was so confident that I thought I knew everything and had nothing left to learn, until the Dunning-Kruger curve slapped me hard enough to wake me up: "You know nothing." That is why learning should never stop.

But now, what I am pursuing is gradually becoming... the opposite. When I run into a problem, I ask AI right away to save time. Once it suggests a solution, I only need to skim it and approve the execution without hesitation. The problem is solved in an instant, but all that remains is: "This model is smart! It knows how to handle this; next time I can just do something similar." Where did those hours spent as the "processor" go? Where are all the trials, errors, and new pieces of knowledge waiting for me to discover them?

Debugging is exactly where I learn the most. It is the perfect combination of searching and reasoning. When an error occurs, solving it forces me to take part in the research process, applying what I know to find a direction. Along the way, I have to search for information or ask others for help. Afterwards comes the step of turning what I have just learned into personal experience. Repeating that process continuously helps build up valuable experience. Yet now, with the development of AI Agents, I do not even need to participate in debugging. All I know by heart is: "Opus 5 can solve this problem within 10 minutes."

Sometimes, when I sit with colleagues, their screens are split into four panes, and all four show the Claude icon. Looking back at my own screen, it is not divided into four, but every VS Code window has a Claude, OpenCode, or even Codex tab. Occasionally, someone is surprised to see me reading documentation or an article online. Google is gradually becoming outdated, while reading itself feels like an extravagance. Well! That makes sense too. Why read when AI can read and do it for me?

Why Can This Convenience Erode Skills?

Cognitive offloading is the use of external tools or physical actions to reduce mental effort and the need to remember. It is like writing something down in a notebook, using a calculator to calculate, or using AI to write or program... The goal is to free up mental capacity, improve efficiency, and reduce stress. That sounds reasonable: people create things to ease the burden on the brain so they can work more efficiently and have more time for other things. But conversely, it can unintentionally weaken memory, reduce critical thinking, and create dependence. Imagine waking up one day to discover that AI models had only been a dream. What skills would remain in you?

Following that, the term Automation Bias has recently been mentioned more often. It is the human tendency to rely too heavily on automated systems and artificial intelligence, because people see machines and software as more objective, authoritative, and trustworthy than themselves. Over time, continuous dependence makes people stop questioning the output. This behavior leads to two main kinds of errors:

• Commission errors: Following an incorrect suggestion or action provided by an automated tool without checking it first.
• Omission errors: Failing to notice a problem or take a necessary action because the automated system does not raise an alert.

Applying this to myself, I realize that I am gradually falling into the trap of both phenomena. If it continues, I may lose my ability to retrieve information and stop spending enough time evaluating reliability. A 2024 study by three researchers, Stadler, Bannert, and Sailer, on ScienceDirect—a leading platform for sharing scientific reports—titled Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry, ran a randomized experiment with 91 German students. One group used ChatGPT-3.5 while another searched the web to prepare a report; the GPT group's argument quality was lower (1.20 versus 1.87). This was not a deep study and only applied to a very narrow sample, but it is enough to show that people were already asking early on whether using AI could limit thinking in some way.

Another study on ScienceDirect, "The Paradox of Information Abundance", was co-authored by Constable, Rajsic, Renner, and Taylor in 2025. Seventy-five participants compared unlabeled answers from ChatGPT-4 and Google Featured Snippets. For "how-to" questions, people considered longer answers generated by ChatGPT more trustworthy and were less inclined to look further, but their recognition and recall scores were lower than with shorter answers from Google. This study once again shows a loss of motivation to investigate a problem while memory declines. Personally, I think this is both a benefit and a drawback; I need to find a balance that lets me make the most of what large language models offer.

That may be why I feel tired, why programming no longer feels as "fun" as it used to. Maybe I have overused it, letting it overwhelm the learning process I once had and dull my curiosity. Or perhaps I simply have not adapted to the new era yet.

The Psychological Paradox: Knowing I Need to Slow Down, but Fearing Being Left Behind

There is no denying the benefits and value AI brings. It is like a massive treasury of knowledge that anyone can tap into, completing in minutes work that once took hours. But if work is being sped up, should that not give us more time to research, reflect, and learn more?

I am not sure! This is subjective and depends on how each person thinks. When everyone can solve work quickly and with very little effort, it accidentally creates pressure: people around me use AI, colleagues use AI, shorter and shorter deadlines become a measure of competence, social media praises those who work faster, and knowing how to use AI becomes an advantage... Refusing to use AI may be seen as deliberately slowing myself down. So when I stand outside it, I feel as though I am moving against the social current.

As an observer, I often read many people's comments on forums, especially whenever AI reaches a new milestone. At first, many people reacted strongly to artificial intelligence, insisting they would never use it or even looking down on AI users. But now, such comments are becoming rarer. If anything remains, it is complaints about privacy, inconsistent quality, or price. Perhaps it is time for them to join the broader trend too, or perhaps they no longer have time to care about these issues. Everyone does their own thing, because what benefits you may not benefit me.

If you spend time browsing social media or joining AI groups, you will see many new products launched every day through nothing but "Vibe coding"—a term for applications created simply by instructing AI. When I click to look, some products are so polished that years of diligent study suddenly feel pointless. Not to mention the endless posts titled things like "How I automated 99% of my programming work." Everyone is chasing automation, extracting every bit of power AI can offer. If I do not learn, do not build, and do not follow the crowd, will I be left behind? When everyone can do the work that was once mine, will there still be a place for me in this unforgiving programming industry?

Ultimately, I think the biggest question I need to answer is this: Is a programmer's sustainable advantage the speed of producing answers, or the ability to ask the right questions, verify them, and take responsibility for the decision?

Do Not Be Afraid; Use AI Like Someone with Their Own Mind

Critical thinking is the process of analyzing, evaluating, and verifying a problem or piece of information logically and objectively, rather than passively accepting it based on emotion or herd behavior. Put simply, do not immediately believe everything you have just received. Instead, ask your own questions or look for the information's true source. I have never felt that critical thinking was more necessary than it is right now.

Over the past one or two years, I have received requests from senior management that use AI more often. A complete AI-generated report that analyzes the cause and state of a system, along with a proposed solution, can seem worth following. At first, receiving reports like that hurt me and made me uncomfortable, as though I were being threatened by every sentence AI produced—so sharp that it was difficult to argue against. Its explanations and logical analysis looked so perfect that simply following them seemed certain to solve the problem. At first, I said nothing, but gradually I began to push back and present evidence. AI can sound convincing, but whether its suggestions can be deployed immediately still depends on the system. Then I discovered what AI does not have: responsibility. Every decision, once applied to a system, still needs someone to take responsibility for it whether it is right or wrong. If I place all my faith in AI, would I not be the one who loses the most?

I need to change how I use AI: set a rule to think first and ask later, instead of asking first and then reading the answer. That helps me return to my earlier way of thinking: search, synthesize, try, fail, get it right, then compare my result with AI's answer to see how they can complement one another. It takes time, but in return it makes my brain think and form a network of connections, while helping me retain more information than one-way intake does. Now, instead of a simple question-and-answer exchange—where I ask and AI replies—I should ask back why it has to be done that way. I am sure there are many interesting things to uncover through this approach.

Even AI providers themselves put a small notice on their products: "AI can make mistakes too. Please verify information before use." So when taking in information from AI, I should always be skeptical of everything it says; it is best to verify it myself. AI can speak very well, clearly, and in a way that makes people believe it at first sight. But from now on, once I have equipped myself with critical thinking, I need to remember not to trust an answer absolutely, and to explore more widely by questioning what AI claims is correct.

Finally, I think everyone should practice the strength of their own mind. Every now and then, try putting AI aside to solve a problem yourself. Think for yourself, write code for yourself, learn a new programming language for yourself, so you can be sure your abilities do not fade with time.

Conclusion

AI helps me work faster, but it also makes me realize that I am easily overlooking the most valuable part of the work itself: the process of searching, trial and error, debugging and forming my own understanding. Reducing cognitive load is not inherently bad, but when combined with automation bias, it can turn convenience into a habit of immediately trusting, immediately acting while forgetting to verify. The pressure to keep up with those who are racing ahead with AI makes me even more afraid of falling behind, but sometimes it is necessary to slow down to reflect on what the lesson learned is? Is it really mine or not?

Therefore, I will still use AI, but I will use it as someone with my own judgment: think for myself first, ask deeper questions, compare the answers with reality and only then act. Occasionally solving a problem from scratch myself may be slow but that is how I preserve curiosity, critical thinking skills and the joy of programming.

This article is only a personal perspective on how to use large language models over the past few years, not a rejection or condemnation although it tends to overemphasize the negative aspects that an AI user feels. Everyone has a different way of using it and I respect that. If you are leveraging AI in a way that brings more benefits and want to share then please leave a comment below the article. Thank you!