Euro
Arrow
55,5183
Dollar
Arrow
50,4140
Sterling
Arrow
66,5689
Gold
Arrow
6286,4579
BIST 100
Arrow
11.699
Home > Authors > Atıf Ünaldı > The AI league and the collapsing software world

The AI league and the collapsing software world

X Facebook WhatsApp Mail SUBSCRIBE Google News
Published: 03.11.2026 06:00

We are debating who is first, second, and third in the world of artificial intelligence. Every week a new model is released, new test results are announced, and someone claims to have developed the world's best AI. Last week, the success Gemini showed in certain tests was the talk of the town. But do these results really show that one AI is better than another?

I think a little differently about this.

The success a technology shows in a laboratory is not always the same as the benefit it provides in real life. In fact, there are often serious differences between them. This is a situation we have seen in the technology world for years. Nothing different is happening in AI.

In the early 2000s, when laptops were starting to become widespread, my friends would often ask me which computer they should buy. I always told them the same thing: Don't buy a computer by looking at laboratory tests. Go, sit down at the computer, put your hands on it, and use it.

Because how fast a computer's processor is, how large its memory is, or how many points it scored in laboratory tests does not mean you will be comfortable using that computer.

For example, some computers Sony sent me for review had nubuck-like surfaces. I don't like touching those surfaces. I prefer smoother, shinier, and if possible, glass-like surfaces. I also don't like the sound the keyboard makes. At one point, I was even using an iPad as my computer's keyboard. Writing silently on a glass surface was much more comfortable for me.

Someone else might prefer the sound a mechanical keyboard makes or the physical feedback of the keys. Is one of us doing it wrong? No. Because technology must serve the user. The user does not have to adapt to the technology.

There is also a printer story I heard years ago.

A company's CEO calls his IT manager and says he wants a very fast printer. The IT department does its research, buys one of the fastest printers on the market, and installs it in the CEO's office.

A few days later, the CEO calls the IT manager again. As the manager approaches the stairs, the CEO throws the printer at him.

Why?

Because the IT department looked at how many pages the printer could print per minute. However, what the CEO wanted was to get a single-page document in his hand in the shortest time possible.

The time a printer capable of printing hundreds of pages per minute made him wait to output the first page eliminated the entire performance advantage from the CEO's perspective.

There are two different performance metrics here: pages per minute and time to first page. Both are technically correct metrics. However, a metric that does not meet the user's need is meaningless.

We are experiencing exactly the same problem in the AI world today.

Why is Gemini not in my top three?

I was recently doing research on quantum computers. I looked for a list showing how many quantum computers are in which country in the world. I asked ChatGPT, I asked Claude, I asked Gemini, I asked Grok. I also used Google's AI-powered search.

The results were interesting.

Some gave the numbers of companies working with quantum computers. Some described the investments countries have made in quantum technologies. Some listed research centers. However, what I wanted was a comparable inventory of existing quantum computers on a country-by-country basis.

None of them could give me exactly the result I wanted. One reason for this is that it is not easy to reach a consensus even on the definition and scope of such an inventory. However, in my user experience, ChatGPT and Claude came closer to the result I was looking for than the others.

This is what AI performance is for me.

Behind Gemini is a massive technology company like Google. It has one of the world's largest search infrastructures, services reaching billions of users, maps, email, video, cloud computing, and countless data sources.

Despite all this, the user experience Gemini provides does not meet my expectations.

I think one of the important reasons for this is a problem Google has not been able to solve for years: its inability to truly integrate its services in a meaningful way.

Google has an extraordinary technology ecosystem at its disposal. But the parts of this ecosystem often act like independent products. I had thought the Alphabet restructuring could overcome this problem. However, as far as I can see, the transformation I expected has not occurred.

Moreover, there are times when I think Google's old voice assistant is more useful than Gemini for some simple tasks. This is another example that technological progress does not always mean progress in user experience.

The same problem appears in Microsoft's Copilot, Apple's Siri, and Meta's AI applications.

You put Meta AI inside WhatsApp. You put it inside Instagram. You are constantly confronting the user with AI. But can it do the job I really want?

For example, I am looking for a conversation I had with someone on WhatsApp months ago. I don't remember who I talked to, but I remember the topic. I want to describe the content of the conversation to the AI and have it find that correspondence.

This is the AI function I expect.

I don't need another chatbot sitting in a corner of the application giving me answers to general questions.

A significant portion of these companies are trying to plaster AI over their existing businesses. Yet, AI is not an ornament added to an existing product, but a technology that should redefine the product itself.

The opportunity WordPress missed

One of the most striking examples of this issue is WordPress.

WordPress has been one of the most important publishing infrastructures on the internet for years. Today, approximately 40 percent of websites use WordPress. This is an extraordinary power.

However, I think WordPress has been focusing on the wrong point strategically for years.

I made a similar criticism when the Gutenberg editor was introduced. Instead of dealing with block-based editing systems, WordPress should have focused on the fundamental issue of internet publishing.

Content.

I never found it meaningful for a publishing system born from the weblog concept to focus so intensely on page design and visual blocks over time.

At a time when page builders like Elementor were successful, WordPress brought a similar approach to its own infrastructure with Gutenberg. But I think this caused WordPress to move away from its true strength.

Now they are facing a much bigger opportunity.

Artificial intelligence.

What WordPress needs to do is create an architecture that can communicate directly with systems like ChatGPT, Claude, Grok, and Gemini through the WordPress core.

Moreover, it should offer this not as a cloud service dependent on WordPress.com, but as an infrastructure that anyone running WordPress on their own server can use.

There are initiatives in this area via Jetpack and various connection protocols. However, these are not the same as the direct core integration I am describing.

With such an architecture, the user would not just have the AI write text. They would manage their site, change its design, edit their content, solve their problems, and control their plugins.

This is the WordPress I want.

The reason initiatives like WPVibe catch my attention is that they approach this need. The user now wants to improve their website by explaining what they want in natural language, without dealing with technical details.

With such a large established user base, WordPress should have been the pioneer of this transformation.

While dealing with Gutenberg, they faced the danger of missing the real opportunity of the AI era.

The skyscrapers of the software world are collapsing

Do you remember the final scene of the movie Fight Club?

Two people are holding hands, looking out the window. The skyscrapers in front of them are collapsing one by one.

When I look at the software world today, I see exactly that scene.

The skyscrapers of the software industry we have known for years have started to collapse.

The economic model upon which Adobe, Canva, and many other software companies are built is undergoing a major transformation. Because a significant portion of users only use a small fraction of the features of the software they purchase.

Photoshop may have hundreds of features. But if my need is only to remove the background of a photo, why do I have to use all of Photoshop?

If I am only adjusting the framing of an image in Canva, why should I remain dependent on a comprehensive application for this?

Now I can describe what I want to an AI and develop a small piece of software that does only that job.

This change will affect not only large software companies but also application stores, software distribution models, subscription systems, and the way software developers work.

I recently experienced this directly while developing my own applications. Preparing, testing, and distributing an application via TestFlight has become a much more accessible process than in the past. Of course, there are still technical problems and platform limitations. But the threshold for producing software is falling rapidly.

What I am saying here is not that people will no longer use software.

It is that the necessity for software to be a product that is purchased, learned, and used for years is beginning to disappear.

Instead of buying software, people will demand the function they need. And the software will be produced to meet that need.

For this reason, I do not believe the traditional software industry can continue in its current form.

And I think one of the AI companies that understands this transformation best is Anthropic.

Why is Claude in second place?

The reason I put Claude in second place is not just the quality of its answers.

I see a direction in Anthropic's approach to AI that could change the existing software order. Especially in coding and software development processes, it is important that AI moves beyond just being an assistant and places itself directly at the center of production.

In this sense, in my eyes, Claude is not just a chatbot, but a harbinger of a new software world.

While traditional software companies are trying to add AI features to their existing products, Anthropic focuses on tools that change how software is produced and used.

That is why Claude's place is a bit different for me.

However, the rise of Grok in third place is also remarkable.

The rise of Grok and the Elon Musk factor

Grok joined the AI race later than its competitors. Despite this, I think it has progressed very quickly.

I am particularly struck by the approach it has put forward regarding AI agents. In a training I recently attended about Grok, it was explained how a company could be established through AI agents.

I think this will be one of the most important issues in the future of AI.

We will not just ask questions to AI. We will give it tasks. It will manage business processes, communicate with other systems, and be able to work on its own within certain boundaries.

Of course, control needs to remain with humans. However, a structure that makes human intervention mandatory at every stage will also limit the efficiency that AI can provide.

Another advantage of Grok is Elon Musk.

When we look at Musk's history, we see that he likes to challenge the business models of established sectors. He did this in financial services with PayPal, in automotive with Tesla, and in the space industry with SpaceX.

At the time PayPal emerged, it was not as common as it is today for people to entrust their financial transactions to a technology company. Tesla moved electric cars to the center of the automotive sector. SpaceX changed the cost structure and competitive conditions of space transportation.

Now he is trying to realize a similar transformation in AI.

The fact that Grok is less hindered by some unnecessary restrictions I encounter in other systems is also an important advantage for me.

I have been saying for a long time that one of the biggest problems with ChatGPT is that its policies on copyright, discrimination, and similar issues sometimes get in the way of the user experience.

I am not debating the legal and social importance of these issues. However, for an AI to spend a long time explaining why it cannot help instead of helping the user is a serious problem for professional use.

I encounter such obstacles less often in my own use of Grok.

This does not mean that Grok is more accurate or more reliable in every subject. But it makes an important difference in terms of the user being able to get their work done.

I see Grok's upward momentum as higher than Claude's. If it continues at this speed, it would not surprise me if the rankings change.

ChatGPT is still the leader but is losing blood

Despite all my criticisms, the number one in my AI league is still ChatGPT.

The reason for this is not just the capabilities of the model. The integrations, tools, and work environment formed around ChatGPT are very important in my daily use.

However, I think ChatGPT has lost its momentum of development.

The fact that it continues to develop technologically does not mean that the user experience is improving at the same speed. In fact, I think it has gone backward in some areas.

Especially unnecessarily long answers, approaches that question the user's decisions instead of helping them, and the transfer of some non-technical restrictions to professional business processes create serious problems.

AI exists to serve a user. Not to make editorial, professional, or personal decisions on behalf of the user.

I see that ChatGPT loses this distinction from time to time.

Therefore, my ranking today is clear:

First, ChatGPT. Second, Claude. Third, Grok.

But this ranking is not permanent.

ChatGPT's momentum is weakening. Claude is rising. And Grok is coming with higher speed.

The results Gemini achieved in tests do not change this assessment of mine. Because what is important to me is not laboratory results, but the benefit that emerges in real life.

Mistral and the future of local AI

Another player that should not be ignored in this race is Mistral.

Important developments are taking place, especially in terms of AI being able to run on smaller hardware and local systems.

I don't get too hung up on parameter counts here. More than how many billions or trillions of parameters a model has, it is important how it uses these parameters, how it relates information, and how meaningful results it gives to the user.

The development that really catches my attention is that powerful AI systems are becoming increasingly capable of running on more accessible hardware.

First, we saw cloud-centric AIs. Now, edge AI, that is, AIs working on systems close to the point where data is produced, is gaining importance. I think the next big stage will be local AI running on personal devices.

Because we need to seriously discuss how much of the information we give to AI should go to other companies' servers.

Why should a company's trade secrets, a journalist's sources, a lawyer's files, or an individual's private life have to be constantly processed in another company's infrastructure?

Local AI is not just a technical development, but also a matter of trust and data sovereignty.

The trust problem China needs to overcome

Chinese AI companies are making significant technological progress. Systems like DeepSeek are among the examples of this.

However, my approach to China-based AI services is not just about technological performance.

When using a Chinese AI service, I don't just feel like a customer of a technology company. I also think that I am a citizen of another country that could have a conflict of interest with China in the future.

This is an important difference for me.

If I don't trust an American technology company, I don't use that company's service. But when it comes to China, my trust problem is not limited to the company. I also take into account the state's influence over companies and under what conditions data can be made accessible to public authorities.

For this reason, a Chinese AI company being very successful technologically is not enough for me to trust that service.

The same approach applies to technological services that could be offered from countries where state control is very strong, such as North Korea.

Of course, American and European companies also need to be questioned in terms of data security. But my risk assessment is different for China-based services.

If I were in the shoes of one of the world's largest technology companies like Apple, I would not want to become dependent on a China-based provider for my critical AI infrastructure.

Even if the cost of this preference is high.

Because in the technology race, trust is becoming as important a metric as performance.

What will we measure success with?

At the end of all these discussions, we actually come to a much bigger issue.

What is success?

When we talk about Elon Musk's success, a significant portion of people immediately say he is one of the richest people in the world.

Yet, money cannot be the only measure of success.

How much money a person earns does not show alone what they have contributed to humanity. Likewise, the market value, investment amount, or the score an AI company gets in laboratory tests does not determine its true success alone.

I think the concept of success itself will change in the coming period.

AI is transforming not only technology but also ways of doing business, companies' economic models, professions, and eventually social values.

Today, we see that the business models upon which massive software companies are built are being questioned. Tomorrow, similar transformations will be experienced in other sectors.

Like the skyscrapers in the final scene of Fight Club, the structures of the digital world that we have considered unshakable for years may start to collapse one by one.

However, what is really important in that scene is not the buildings that are collapsing.

It is the two people looking out the window.

Because it is still up to people to decide what kind of world will be built after the order that has collapsed.

Who will be first in the AI league also depends on this question in the end.

Not who developed the biggest model, but who really made human life easier?

And perhaps more importantly, who was able to protect the human's control over their own life while doing so?

Author's Latest Articles All Articles