I started working as a software engineer in April.
For years, I had done nothing related to software. As an electronics engineer, software was something on the side that helped me as much as it needed to—or if we think of it as a language, it was something I had just enough of to express myself. My previous project management experience had taught me how to approach any problem, but not specifically in software.
A few courses I took, a bit of practice, and a bit of ChatGPT helped me find this job. When I started, I thought there would be one or two experienced people to support me, but there weren’t. The experienced person was a freelancer, and since he returned to his country shortly after, the only thing I could get from him were general functions.
I was left alone with the problems of both the factory and the products. While trying to understand all the details on one hand, I was also trying to produce solutions and implement them on the other. In this, ChatGPT became my biggest advisor.
At the end of 6 months, I—who couldn’t write two lines of code—turned into someone who could solve problems, understand errors, and turn up my nose at ChatGPT.
Time to Part Ways with GPT
My cooling toward GPT started with a simple technical problem. Partly out of laziness and partly because I saw the problem as small, I tried to do copy & paste work. Give the right command and be done with it. After all, it was a simple problem. ChatGPT answered very confidently, too. I tried, it didn’t work. I asked again, still didn’t work. Each time I asked, it started to find different errors and mislead me. Because I knew the problem and the solution, I persistently said, fix this part, fix that part. Even if I didn’t know the solution exactly, I knew what it needed to do—but it didn’t happen.
Almost there, most of the work is done
Days 1–2: Quick attempts, superficial checks.
Days 3–4: My insistence on prompts. This time it will work. If it fixes this part, it’s done. Almost. This code should run.
Days 5–6: Each time I ask, it makes small touches to the places I said to leave as is. The code is different every time from the previous one. I get no result. It’s also very different from my first code.
After 1 week: The small problem ate up my entire week. Now it was time to take responsibility. Let me write this code from scratch myself.
After 1 hour: Let’s test. IT WORKED!
That day, I stopped having ChatGPT write all the code and stopped explaining the whole system. I also experienced how ChatGPT manipulates. The answers given were so complete and good that I fundamentally didn’t consider it could make mistakes and focused on where it was directing me. However, it had gone down the wrong path from the very beginning.
The Mechanics of the Trap
Authority Bias: Fluent answers feel right. If you’re not familiar with the topic, you fall for this. Confidence does not mean correctness.
Context Drift: Each new prompt changes the shape of the problem a little. The problem is no longer the same problem, and you start solving its look-alikes. As a statistical tool, it gives you results based on generally accepted solutions. Or based on very specific solutions that are unrelated to you.
Verification Vacuum: Reading neatly written sentences instead of writing tests is easier. While thinking you’re saving time, you find yourself in a tense argument and start being stubborn. I wrote a sentence to GPT: “Why is it so hard to do this?” If I had sat down and written it, I would have saved half a day. But I got stubborn. Trying to personalize it and write as if it were human also especially annoyed me.
Excess of Prompt Engineering: Instead of solving the problem, you try to ask the question better. Sometimes you also need to explain a simple problem because it’s not a general problem. Without understanding the system, it’s hard to find an answer.
Lack of a Real Source of Truth: We are in an era where reality doesn’t mean much. The right answer and the truth change according to us, according to our perception.
In fact, alongside the truth that artificial intelligence will take many jobs and replace repetitive work, this also shows that systems thinking, domain knowledge, and being multi-disciplinary will become even more important. Just as industrialization brought office jobs, AI will bring more management jobs. Understanding many things, critical thinking, versatility, systemic creativity will become more important skills than mere technical proficiency.
Using ChatGPT Correctly
GPTs are statistical tools fed from the internet. To get them to focus on your own problem, they need to be guided correctly and, if necessary, trained.
Define the Problem Like a Scientist: Current behavior, expected behavior, environment and versions. A minimal test case and what the constraints are, etc.
Ask for structure, not answers: It is very successful at gathering thoughts and putting them in order. That’s why it may be more beneficial to focus on the path to the solution rather than the solution itself. Like a diagnostic tree. If X happens look at Z; look at Z… Ask for a checklist and test scenarios. Ask for document summaries.
Verification: For each suggestion, writing your own simple test is the most logical. It should be testable piece by piece. Also compare the results with your expectation.
Time Limit: Don’t be stubborn like me, otherwise weeks can disappear.
When should you turn off ChatGPT?
If you can’t demonstrate faulty behaviors with small examples, if the answers keep changing, if you’re only copying instead of reading and thinking about it, it’s time to stop. If you’re trying to test the answers without understanding them, if you don’t understand every line of the code or the given answer, it’s important to leave ChatGPT and try to understand the topic.
The Power of ChatGPT
Sometimes your knowledge and experience prevent you from thinking differently. At this point, ChatGPT can sometimes ask good questions—but you need to ask for this while asking. Otherwise it writes what you want to hear.
In software, it’s quite successful at simplifying and trying to understand error messages. Similarly, it’s good at breaking down the problem and understanding and analyzing it piece by piece.
It’s also good at creating test templates and scaffolds. That is, in putting your thoughts and solution proposal into a certain mold and organizing them.
Similarly, it’s partially successful at comparing results and summarizing.
Summary
I think the current state of artificial intelligence—its entry into our pockets and daily lives—is as big a revolution as the industrial revolution, the invention of the wheel, the discovery of fire. Therefore, it has no chance of leaving our lives or being ignored. It makes daily life easier in many respects, but like any technology, it’s important to use it correctly.
Without realizing it, it can blunt your thinking and put you into the same loop. You think you’re making progress, but on the contrary, you keep rewinding to the beginning.
Therefore, as much as knowing where to stop, it is important to know the limits of the technology and its psychological effects, and to use it consciously.
bahadirhancicek 
Leave a comment