What we went through with AI in 2025, in Turkey and globally

In relative terms this was probably one of the fastest and most transformative years for all of us. We left a lot behind, the good and the bad, but what I want to focus on here is this: looking back at all that speed, what happened in the IT sector during 2025, and how did it reach us in Turkey. I did not use the word llm anywhere in this post, and for the picture I used a landscape photo I took in the village earlier, so we can all breathe a little.

I think it was January 2025, during a short spell I spent abroad, when a friend came to visit. He said things like “there is this thing called Cursor, you would really like it.” To be honest I did not take it very seriously at first. I would sit down, draw the database schema in dbdiagram, then design the api endpoints, then start coding, I told him. Because Copilot and the rest never produced output I found satisfying, they slowed me down instead.

Then one day I decided to look, and what I found shocked me. This tool was giving frontend support to someone who had worked as a senior backend developer, and it removed my need for a frontend partner if I wanted to build the MVP of a project. That day was the revolutionary change of AI in my life. From then on I tried to build an MVP almost every day. Frontend, backend, the devops side, and finishing all of it took me twelve or thirteen hours for the MVP of a serious project.

We started following the models. With every new model our trust in AI grew. Sonnet 3.5, Sonnet 4, and then the others came. We had the whole DeepSeek moment. We also saw that the thing we call AI might not be an American monopoly. With the latest models such as Gemini 3 and Claude Opus, it has reached a point where it is trustworthy enough for coding.

So how did companies react to all of this? When 2025 started, I asked every colleague I was in touch with in Turkey and learned that companies had taken almost no action on the subject. At first it felt like a big case of ignoring it, but while ignoring it is possible as an individual move on the developer side, on the company side it was less about ignoring and more a state of inertia that came from not having seen it coming. When I showed people the things I could do as a product manager after work hours, they were quite shocked. Companies in Turkey noticed the situation more towards the middle of 2025. Around June and July, many of them reacted strongly and started making big, radical moves. Firms and individuals who advise C level executives pushed them hard with the message that they had to take action on AI.

The biggest mistake made here was that companies, under the banner of “we must use AI”, moved very fast without building any strategy, and leaned towards training that touched more people in volume and output that was larger in volume rather than towards quality training and quality output. Companies in Turkey could not think about ROI on AI in this sense for a long time. The main reason is that AI was understood only through generative AI and through the currently fashionable idea of agents. In reality many companies in Turkey have had data science departments for a long time, and serious work on machine learning and related subjects is done there. But with PR driven moves and the perception that they had to be a leading company on AI, companies turned mostly to generative AI. In that context a lot of work went into media production during 2025.

Yet at the start of 2025 image generation with generative AI was nowhere near functional, and only towards the end of the year did it start becoming suitable for corporate use. I am not even talking about the ROI side of it. Right now managers in companies are deep in budget planning because the year is ending, and on a subject where they mostly have no idea, they have serious doubts about how much to invest and how to plan.

On the international side, the same worries exist at leading companies, but on the operations side I can say they put more weight on quality than on volume. From the news I read and the developer blogs I follow, work abroad puts far more effort into stability after AI agents. Standards have been built for the stability of systems, and agents that do not meet those conditions and do not follow those regulations are not put into production.

2025 pushed many companies on vision and on judgement. It pushed their vision first, and the companies that started the transformation early became the leading ones. It pushed their judgement too, and companies that made unnecessary investments and planned badly with AI strategies not grounded in knowledge met, and will meet, very unnecessary costs. From now on companies and managers should think through every step not twice but ten times. They should follow the present far more closely. We saw how far the conservatism of the companies we call giants set them back in this period.

My expectation for 2026 is that what happened in 2025 comes back far more intensely. I think we have reached a very good point in coding with LLMs, which does not mean it will stop improving. I think we will see more developments in visual and audio production, and in robotics, during 2026.

I wish everyone a peaceful working year.