This week I read an article on CSS-Tricks about Stack Overflow. It showed a chart, and the chart stayed with me. Questions on Stack Overflow peaked around 2014, with more than 200,000 questions in a single month. In 2026 the site barely reaches 3,000 a month. The author points at AI as the main reason, mostly.
I felt something when I saw that drop. I remember asking questions there years ago. I remember the small fear before posting: maybe my question is a duplicate, maybe someone tells me it is stupid. But I also remember how much I learned from that process. Writing the question forced me to understand the problem first.
Now I almost never ask Stack Overflow. I ask an LLM. It is faster. It does not judge me, in fact it praises me too much. I get an answer in seconds. And honestly, I like it.
But the more I think about it, the more I worry. My worry is not really about Stack Overflow. It is about something bigger. AI is changing how we think, how we write code, and how we read it. I am not sure this goes in a good direction.
How we think
There is a word for what happens here: cognitive offloading. We move mental work to an external tool, so our own "muscle" does less. Sometimes this is fine. We use calculators and we do not feel bad about it.
But recent research suggests a cost when we offload too much. A 2025 study by Michael Gerlich looked at 666 people. It found a negative link between heavy AI use and critical thinking. The effect was stronger for younger people. The author calls one part of this "cognitive laziness", a drop in the wish to think deeply and reflect.
A study from MIT Media Lab goes further. Researchers asked 54 people to write essays in three groups: one used an LLM, one used a search engine, and one used only their brain. They measured brain activity with EEG. The brain-only group showed the strongest and widest brain connectivity. The LLM group showed the weakest. The detail that scared me most: most people in the LLM group could not quote a single sentence from the essay they had just written. The text was there. The thinking was not.
This matches my own feeling. When I let the model do the hard part, the answer does not really stay in my head. I ship it, and a week later it feels like someone else wrote it.
How we write code
The same pattern shows up in the code itself. GitClear analyzed 211 million changed lines of code from 2020 to 2024. The trend is clear and not great. Copy-pasted lines went up from 8.3% to 12.3%. "Moved" lines, which usually mean refactoring and reuse, dropped from around 25% to under 10%. In 2024, for the first time, developers pasted more code than they moved.
So we add more and reuse less. We duplicate instead of cleaning up. Research from Cornell points in the same direction: AI-generated code tends to be simpler and more repetitive, with more unused parts. It works, but it does not always make the codebase better.
For someone who likes boring, standard Rails, this is the part that bothers me. Good code is not only code that runs. It is code that stays simple over time. Refactoring is how we keep it that way. If we stop refactoring, the system slowly rots, even if every single pull request looks fine.
How we read code
There is a third change, and people talk about it less. AI is changing how we read code, or how little we read it.
When the model writes a function, the easy move is to accept it. It looks reasonable, the tests pass, we move on. But reading carefully is real work, and AI makes it very easy to skip that work. Psychology Today describes a study where developers who handed coding to AI produced working code but failed to understand it on a conceptual level. The code shipped. The understanding did not.
This creates a strange position. To review AI output well, you need the skill that the AI is replacing. You need to know what good looks like. A senior developer can audit the output, because they built that knowledge over years. A junior who started with AI may never build it. You cannot check work that you were never able to do yourself.
I am not against the tool
I want to be clear. I use AI every day and I am not going back. The tool is useful, and refusing it would be silly.
But I try to keep some rules for myself. I ask small, specific questions instead of asking the model to build the whole thing. I read the output and ask if I really understand it. I check where the answer comes from. And I test it like I would test my own code, because the model does not know my users.
These rules are simple, but they have one goal. I want to stay the person who thinks, not the person who only approves.
The real question
Stack Overflow had a lot of problems. The moderation was harsh, and beginners often felt unwelcome. I do not miss that part. But the site did one good thing for many years. It got us to ask. It got us to answer. It got us to think.
The question I keep coming back to is simple. When we stop asking, do we also stop thinking? And if we hand that thinking to the model, who is left to teach the model when the next thing changes?
I do not have a clean answer. I just know I want to keep asking questions, even the ones a machine could answer for me.
So I want to ask you. Do you think AI is taking away our ability to ask questions and to think? Or is it turning into something else, a new way of thinking that we do not fully understand yet? I would like to hear how you see it.
References
- Sunkanmi Fafowora, "Stack Overflow: When We Stop Asking", CSS-Tricks (2026)
- Nataliya Kosmyna et al., "Your Brain on ChatGPT", MIT Media Lab (2025)
- Michael Gerlich, "AI Tools in Society: Impacts on Cognitive Offloading", Societies (2025)
- GitClear, "AI Copilot Code Quality 2025"
- "Analyzing the Differences Between Human-Written and AI-Generated Code", Cornell (2025)
- "Adults Lose Skills to AI. Children Never Build Them.", Psychology Today (2026)
I write about Ruby, Rails, and software development every Thursday. You can follow me at enderahmetyurt.com.