AI is probably a word almost everyone has heard by now in 2026. But for me, it only started feeling real near the end of 2022, when ChatGPT was released. Before that, I mostly heard terms like data science and data analytics. The word AI still sounded like something much bigger and more distant to me at that time.
It was during the time I was pursuing my second Bachelor's Degree in Computing at KMD.
I had just finished one of my small mobile projects written in Java. I lay down on my bed, started scrolling through YouTube, and came across some content about ChatGPT.
I didn't know much about ChatGPT, but I already knew OpenAI because of OpenAI Five, the Dota 2 matches between professional human players and AI, and the popular 1-vs-1 mid-lane matches between AI and human players a few years earlier.
Anyway, I created an account and tested ChatGPT by asking some basic questions. I hadn't asked anything related to coding yet. I mostly asked simple questions about things happening in Myanmar, and it would often tell me that its training data only went up to September 2021.
Then one of my friends, who had started studying Python, showed me that he could simply dump error messages into ChatGPT and get back surprisingly useful responses, almost like asking another developer for help.
That was when I knew this ChatGPT thing was going to become my self-study mentor buddy.
AI at that time was nowhere near as capable as what we have now. It was mostly a simple question-and-answer chat interface with a few options, and many other AI websites were basically ChatGPT wrappers.
One of the biggest limitations was simply getting access to it. ChatGPT gained so many users within a couple of months after its initial release that, during traffic spikes, I sometimes had to wait for hours before I could use it again.
Still, it completely changed how I learned programming. I rarely had to dig through Stack Overflow just to understand an error because ChatGPT could usually explain what was going wrong and help me work through it.
But there was an obvious downside: its knowledge could be outdated. One example I remember clearly was Next.js. Around that time, Next.js was shifting from the Pages Router toward the App Router in version 13, while ChatGPT would often give me solutions based on the older Pages Router even when I was trying to use the newer App Router.
That was my first real experience with one of AI's weaknesses: it could sound completely confident while giving you an outdated or simply wrong answer.
A couple of years later, when I started my Master's in Computer Science in 2024, my relationship with AI changed again. By that point, AI had become much more integrated into the software engineering ecosystem. I already knew how to ask it questions even without writing quality prompts, debug code with it, and recognize when it was probably hallucinating.
But I still did not really understand what was happening underneath.
If you are the artisan, at least you gotta know the tool you are using, right?
So during my studies, I focused much more on Machine Learning and AI-related classes and gradually developed a better intuition for how these models are trained, why they behave the way they do, and how different architectures work.
Learning those things also helped me understand the scale of the AI systems we casually use today. Frontier AI models can have hundreds of billions of parameters, and some architectures reach into the trillions. They are trained on enormous amounts of data and backed by equally enormous infrastructure.
Because of that scale, even general-purpose chat models with enough context and a good prompt can already give surprisingly strong answers across many areas of programming, sometimes at a level that feels like talking to a very experienced developer.
Now imagine systems designed specifically around software engineering workflows, such as Claude Code or Codex: coding agents that can inspect repositories, edit multiple files, run tests, debug failures, review code, and keep working across longer tasks.
There's no point pretending an individual developer should try to compete with these statistical pattern-matching giants at the things they are naturally good at: generating large amounts of code, recalling syntax, searching through huge contexts, or repeating a task hundreds of times without getting tired.
This is very different from what most of us were doing back in the chat-based AI era of 2023, where the workflow was basically: ask a question, copy some code, paste an error back, and repeat.
Modern agentic systems go much further than that. I am not talking only about generating code anymore. They can reason through tasks, use tools, inspect their own results, detect failures, iterate, and try again with much less human intervention than before.
This also means the entry barrier for building software has become much lower. Non-programmers and low-code or no-code users can now build websites, small applications, internal tools, and automation workflows much more easily than they could before.
And that is where the uncomfortable part of the discussion begins.
If AI can generate code, debug problems, use development tools, and even iterate on its own work for hours, then what happens to human junior developers, mid-level developers, and new graduates entering software engineering?
So... do we just give up, ditch our jobs, current studies, and go farming?
NO!
For Junior Developers
I recommend using AI as much as possible every day. Not for blind copy-pasting or massive code generation, but for discussions during your free time and for asking about things you genuinely want to understand.
Remember, to become a good developer you need to stay curious and, most of all, consistent. Try to invest at least a few focused hours every day in hands-on development. Not just reading tech articles or watching YouTube tutorials, but actually building things, breaking things, debugging them, and figuring out why they work.
During that time, use AI carefully as a mentor. Let it explain the code it generates, ask why certain decisions were made, and spend time reading and understanding enough of that code to build your own intuition. You do not need to understand every single line immediately, but you should understand the important ideas and how the pieces fit together.
If you still don't understand something, go back to the basics or
isolate that specific part in a small separate repository and
experiment with it on its own. I find that this gives you a much
higher chance of actually understanding what is happening instead of
just staring at a larger codebase. I still do this myself whenever AI
generates something that goes beyond my current understanding.
I also wrote a practical guide about how I personally read and
navigate AI-generated code in VS Code. You can read it
here.
Don't be afraid to think for yourself just because AI can already do many of these tasks. Form your own opinions based on what you have learned so far, then try to apply that knowledge together with AI instead of letting AI make every decision for you.
Perfection kills. AI-generated code, websites, or other things can look flawless at first glance, but sometimes you can still feel that there is no real soul or craftsmanship behind them. And ridiculously, people can often detect that too, even if they cannot exactly explain why. Spend time making things your own. Rewrite parts, experiment, make mistakes, and develop your own taste.
This is where your own cognitive skills start becoming much stronger. When you spend hours actually working through a project, you start seeing the next critical steps yourself instead of always waiting for AI to lead you there.
And one final thing I want to mention: as of mid-2026, AI still does not have real-world embodiment in the same way humans do. It can do wonders digitally, but in the real world you still have a huge advantage through working with real people. So do not ignore your interpersonal skills. Learn how to communicate, explain your ideas, work with others, and build genuine connections.
To be continued...