A year building with AI. Why did he have to start again?

For a year, he built an application of his own. The problem was that artificial intelligence wrote the code for him. Eventually he wanted to change something and could not: the program stopped working, and he no longer understood what was happening inside it.
Mateusz Bogolubow, founder of DevMentor.pl, shares the experience of one of his mentees in an EdTech Dots conversation. He says he is simplifying the story, but its conclusion stayed with him: the learner had to return to the fundamentals. The application was taking shape, while the skills needed to work on it had not kept pace.
The story raises a question: who does the work that learning requires?
Try the problem before asking for an answer
Bogolubow suggests spending time on a problem yourself before turning to AI. If you get stuck, ask for a first step or a hint instead of a complete solution. The approach takes patience, but it leaves room for your own decisions and mistakes.
He refers to the idea of productive failure: encountering a difficult problem before seeing the solution can be part of learning. In his account, the point is not struggle for its own sake. It is the effort to understand the task before someone shows you an answer.
In practice, the change can be small. Instead of pasting in an instruction and copying the result, try describing what you already know, what you do not understand and what step you might take. Then ask the tool to help you examine that reasoning.
The code works. Can you evaluate it?
Bogolubow expects programmers to spend more time reviewing AI-generated code. He does not conclude that they can stop learning to code. To assess a result and improve it, he argues, they need to understand what it does.
He compares this to language learning. You may understand what someone says and still find it difficult to express an idea yourself. Likewise, receiving a working piece of code does not necessarily mean you could solve a similar problem when the conditions change.
That is one of the questions he would put to a candidate: what problem did the application solve, and how would you approach a similar but different one? The aim is to understand the work behind the project, not simply to see it in a portfolio.
A mentor does not have to fix everything at once
One observation in the conversation stands out: a good mentor should also know when not to point out a small mistake. Bogolubow explains that a beginner who receives fifty comments on a first code review may take away a single message: that they are not cut out for the work.
In his own mentoring, he tries to address the most important issues first and return to others later. Remembering where a learner struggled before, and whether they are making progress, also matters.
This does not mean AI cannot explain things. His concern is about how people often use it: they seek a quick solution instead of asking why the answer works.
How do you find out whether programming suits you?
When someone considers changing careers, Bogolubow first asks why. Good pay can spark interest, but before leaving a current job he suggests finding out whether the work itself is enjoyable.
His proposed test is straightforward: open a free course, write code alongside the instructor, then change it and watch what happens. Even early experiments with HTML and CSS can show whether you enjoy working out why something behaves as it does. He also sees a place for AI as a tool that can spark curiosity and encourage someone to look inside an application.
The proposed test starts with a small task and your own changes to the code. AI can help, but one question remains: can you explain why the changed program works?
This article is based on Christopher Kosman's EdTech Dots conversation with Mateusz Bogolubow.



