Why I Decided to Write This Now
I think the time has finally come for me to write about artificial intelligence. More specifically, I want to write about how I use it in my daily work, how I see the changes happening around us, and why I decided that this is exactly the right moment to publish these thoughts.
That last question is actually more important than it may seem. I've wanted to write about this topic for a long time, but every time I started organising my thoughts, I felt that something was still missing. I knew that the world was changing, I knew technology was moving incredibly fast, but I couldn't yet connect everything into one coherent picture.
Eventually I realised that trying to fit everything into a single article simply wasn't possible. There were too many observations, too many conclusions, and too many questions that deserved their own discussion. Instead of forcing everything into one long story, I decided to split it into several articles, each representing one point on the same journey.
The first article is about where we were and where we are today. The second one will focus on where I believe the IT industry is heading over the coming years. Finally, the last article will zoom out much further and discuss what this technological direction may ultimately mean for humanity as a whole.
Until quite recently I still wasn't confident enough to write any of this. I understood that enormous changes were coming, but I couldn't clearly see their shape. The direction existed somewhere in my head, but the lines connecting everything together were still blurry.
This year, however, something changed. Not because a single revolutionary breakthrough suddenly appeared, but because enough practical changes accumulated for me to finally recognise what I believe is the second important point on this journey. Once that second point became visible, it suddenly became possible to draw a line between the beginning and today.
Of course, drawing a line doesn't guarantee that every prediction will come true. The future always contains uncertainty. Some things may happen sooner than I expect, others much later, and some ideas may prove to be completely wrong. Nevertheless, I believe that the overall direction has already become visible enough to begin talking about it.
The Four Dots
Every forecast, regardless of whether we're talking about economics, technology or even everyday life, starts with the same simple principle. To predict any direction, you need at least two points. Without them, there is no line, and without a line, there is no meaningful projection into the future.
The first point tells you where you started. The second point tells you where you are today. Only then can you begin estimating where the third point might appear, and perhaps even imagine a fourth one.
For more than three years, that was exactly my problem. I had the first point very clearly in my mind, but I couldn't identify the second one. Something significant was obviously happening, yet I couldn't confidently answer whether we were witnessing another technological trend or the beginning of something fundamentally different.
Looking back now, I think 2026 became that second point for me. Not because everything suddenly changed overnight, but because enough progress accumulated to make me stop and think: "Yes, this is no longer just an experiment. This is already a meaningful checkpoint."
Once that happened, drawing the line suddenly became much easier. That's exactly what this first article is about. We'll look at where everything started, where we are today, and what happened between those two points. Only after understanding those foundations does it make sense to begin talking about the future.
My First Contact with AI
For me, the story really begins in 2022. That was the year when I first encountered ChatGPT. Perhaps it appeared a little earlier, but like many other people, I truly discovered it somewhere around the middle of that year.
I still remember my very first reaction. The first thought that came into my head was surprisingly simple:
"These answers must have been prepared in advance."
That was honestly my first impression. Alongside that thought came another one: this simply couldn't be real. Producing responses like that felt far too difficult. It looked more like science fiction than software engineering.
Looking back today, I think what OpenAI accomplished wasn't some magical discovery that nobody had ever imagined. Rather, they were the people who believed strongly enough in an idea to carry it all the way to completion. Many researchers probably considered similar concepts before them. The difference was that somebody finally proved that the idea could actually work.
The general concept itself wasn't impossible for me to understand. If you analyse billions of sentences, calculate the probability of one word following another, and repeat that process on a massive scale, it becomes plausible that a machine might eventually predict the next word in a sentence. At least theoretically, that approach seemed reasonable.
The real question wasn't whether such an algorithm could exist. The question was whether it could work well enough to feel natural.
Why Language Models Surprised Me
To understand why ChatGPT impressed me so much, it's important to understand what I expected before I started using it.
Imagine someone asks a very simple question:
"What is the capital of France?"
If a model predicts the word Paris after the word France, that doesn't seem particularly extraordinary. That relationship exists in millions of books, articles and websites. Given enough training data, a machine should eventually learn that association.
But then consider a completely different example.
Someone simply writes:
"What is the capital..."
At that point there isn't enough information to answer the question. No country has been mentioned yet. Should the model continue with France? Germany? Italy? Or something else entirely?
What surprised me was not that the model guessed another word. What surprised me was that it understood the question itself was incomplete.
Instead of trying to predict a country, it could respond naturally with something like:
"Could you please specify which country you mean?"
That completely changed my understanding of what was happening. I had expected sophisticated word prediction. Instead, I saw something that looked much closer to an actual conversation.
At that moment I realised that the amount of knowledge available for training had finally reached a point where a model could maintain context, ask clarifying questions and guide a discussion in a surprisingly natural way.
Looking back today, this may seem obvious. At the time, however, it genuinely felt like crossing an invisible line. Something that previously appeared unrealistic had suddenly become practical.
Machine Learning Before ChatGPT
Artificial intelligence itself was not new to me. Long before ChatGPT appeared, I had already worked with machine learning concepts and had followed their development for many years.
My first real exposure to this field goes back to around 2013. Even then, machine learning models were already solving surprisingly difficult problems. Companies like Google and Yandex were using neural networks to recognise images, classify objects and extract information from photographs.
A computer could analyse a picture and determine whether it contained a car, a bicycle, a shirt or a person. It could distinguish patterns, colours and textures. It could even recognise details that, from a technical perspective, are far more complicated than they initially appear.
We rarely appreciate how difficult vision actually is because our own brains perform it effortlessly.
A person walks into a kitchen and immediately understands the environment. We know where the walls are, where the table stands, where the door is located and how to move through the room without colliding with anything. None of this requires conscious effort.
Teaching a machine to perform even a fraction of those tasks is remarkably complex. Yet more than ten years ago, computers were already becoming surprisingly good at recognising visual information.
Because of that background, image recognition never felt particularly futuristic to me. It was an impressive engineering achievement, but it was still relatively easy to understand how such systems could evolve over time.
Language was different.
Human language is full of ambiguity, incomplete thoughts, context and implied meaning. Predicting a single word is one thing. Maintaining a meaningful conversation is something entirely different.
That was why conversational AI impressed me much more than image recognition ever had.
My First Experiments
Naturally, after discovering ChatGPT I began experimenting with it in my daily work.
Like many software engineers, my first instinct was simple: could this become a junior developer?
I tried multiple times over the following months. Every six months or so I would return, ask similar programming questions and compare the results with my previous experience.
My conclusion at that time was fairly straightforward.
It wasn't a junior developer.
At least not yet.
Instead, I started looking at it from a completely different perspective. Rather than treating it as another engineer, I began thinking of it as an exceptionally well-indexed version of the internet.
That distinction is important.
Traditional search engines return links. You still have to decide which result to open, compare multiple sources, filter contradictory information and build your own understanding.
ChatGPT worked differently.
It analysed the question itself, understood the context significantly better than a traditional search engine and presented information in a way that already resembled a conversation rather than a collection of hyperlinks.
From my perspective, that was its real value at the time.
It wasn't replacing software developers. It wasn't writing entire production systems on its own. But it dramatically improved the way I searched for information, explored new topics and learned unfamiliar technologies.
Even in programming, where I originally expected the biggest revolution, the practical improvement was relatively modest. Looking back at that period, I would estimate that it increased my coding productivity by roughly ten percent.
That may not sound particularly impressive, but something much more significant was happening in parallel.
Outside of programming, my learning speed increased dramatically.
Whenever I wanted to understand a new technology, research a concept or explore a completely unfamiliar domain, I no longer needed to spend hours navigating dozens of websites. Instead, I could have a conversation, ask follow-up questions and continuously refine my understanding.
That turned out to be a much bigger productivity gain than simply generating code.
A New Way to Search for Information
Over time, I gradually introduced ChatGPT into my daily work. It wasn't a conscious decision to replace anything. It simply happened naturally. Around six months after I started using it, I realised that I had almost completely stopped using Google.
The only time I still opened Google was when I wanted to find a specific website. In most browsers it is still the default search engine, so typing the name of a website and opening it directly remained convenient. Apart from that, I hardly used it anymore.
At some point I remembered something I had read years earlier about Google. Whenever you searched for something—for example, "Titanic"—Google displayed the number of results that supposedly matched your query. I remember reading that these numbers were not always exact. Whether there were one hundred thousand pages or one million pages didn't really matter. The number itself had more of a psychological effect than any practical meaning.
Looking back today, I occasionally still open Google. What I noticed is that I almost never go beyond the first page of results. Quite often, even the first page no longer feels particularly relevant to me.
I don't think this necessarily means that Google became worse. Rather, I think the way I search for information has changed.
When you use a traditional search engine, you open one website, read part of the information, then open another website because it contains something the first one didn't mention. A third website presents the same topic differently. Every page has its own layout, its own structure and its own way of presenting information. As a reader, you constantly have to adapt.
That adaptation requires effort. Your brain has to switch from one source to another, compare information and decide which parts are useful. Eventually you build your own understanding by combining information from many different places.
With ChatGPT the experience is different.
Instead of searching through individual websites, you work with information that has already been aggregated. The model understands your question, provides an answer, asks follow-up questions when necessary and continues the conversation within the same context.
You no longer need to adapt to the way different websites present information. The conversation remains consistent while the information changes.
That was how I started looking at it. For me, ChatGPT became a new form of search rather than a replacement for software engineers.
What I Saw During My First Months
When I first started working with it, I tried to use it as a junior software developer. I repeated that experiment several times over roughly six-month intervals, hoping that it would eventually reach that level.
At that time, however, I couldn't achieve the result I wanted.
My impression was that, in software development, it helped me, but only to a certain extent. If I had to estimate it, I would say that it improved my productivity by around ten percent.
The much bigger improvement happened somewhere else.
It significantly accelerated my learning process. Whenever I wanted to understand a new technology or explore a topic outside my everyday work, I could simply ask questions, continue the discussion and gradually build my understanding without constantly searching through dozens of different websites.
That was the point where I started treating ChatGPT differently. I no longer expected it to become another developer working alongside me. Instead, I treated it as a completely new way of finding and working with information.
Looking back now, I think that was where the real change began—not because it could write code, but because it fundamentally changed how I searched for knowledge and how I learned new things.