AI-driven world: 2025

A quarter of a century has passed, during which we have found ourselves in the pages of science fiction books. In 2000, we rejoiced at the first mobile phones, but did not understand why a phone needed a camera, or why we needed rewritable DVDs - give us two, please. The first 32MB flash drives were cool storage devices that fit in your pocket. A little later, “mobile PCs” like the Palm caused surprise, and for most people, the Internet was still more of a geek's entertainment than a basic necessity.

And now, just 25 years later, we live in a world where the Internet is sometimes more accessible than electricity. Everyone has a device in their pocket with a capacity unimaginable in those days, constantly connected to cloud services. It has also become possible to explain to a machine what to do in a language familiar to anyone, and it will do it.

AI technologies have become commonplace in just a few years, but in terms of their significance for humanity, they probably surpass the advent of the Internet, which, incidentally, took decades to spread.

Below, I have described my thoughts on this new world for a New Year's post to read over a glass of something.

1. AI will (not) replace programmers, lawyers, writers, marketers, and anyone else.

This is the most popular debate of the year. It is surprising how polarized people's positions are on this issue. Some categorically state that profession N has two years left. Others show examples of how ChatGPT generates incorrect code and just as categorically assert that AI is useless and will never replace professionals.

It's not clear what the debate is about. Even if LLM takes on 30-50% of the tasks, say, writing code or processing information, that's already a huge increase in work efficiency. Although for a wide range of expensive tasks, these tools take on much more.

Will AI replace lawyers? Not in the near future, at least not with all the regulations and bureaucracy. But the speed at which they work could increase significantly, and high-quality legal services could become more accessible.

Will AI replace programmers? Most likely not. We used to write everything ourselves, then libraries and frameworks appeared, freeing programmers from low-level routine tasks. The code layer is becoming less significant, but it is mainly template code that has been written millions of times by millions of people. Now you don't have to listen to explanations and wait weeks to add a button to an application, even if there is complex logic behind that button. Programmers are becoming more like architects and can focus on important and interesting tasks.

2. Fundamental knowledge is becoming more important

Many craft skills can indeed be replaced by AI. Now it is more important to understand the essence of things, how they work, how they interact, and why. Specialization in applied knowledge is becoming less valuable, as knowledge is available almost instantly. However, a systematic approach to problem-solving in any field will be the essence of professions where AI can be applied. Understanding fundamental principles and the ability to put together a ready-made solution are now a necessity.

This raises the entry threshold; for a huge range of “junior” tasks, people are no longer needed. Every specialist becomes an architect, analyst, engineer, and manager. I believe this is also true for lawyers, accountants, marketers, game designers, and possibly even scientists.

3. Strike while the iron is hot... and cheap

Most consumer AI products are now very affordable. I would say they cost pennies compared to their value. All these Claude Code, Cursor, and others for a couple of dozen, even hundreds of dollars a month, are nothing compared to the hundreds of man-hours they replace.

However, behind this are huge expenses for hardware, venture capital, and a period of industry saturation. Sooner or later, prices will rise to recoup these investments. I predict that in the next three years, they will increase fourfold. So this is a reason to squeeze the most out of what is available in 2026.

4. Hardware shortage

There is a looming hardware crisis for growing models, RAM prices are rising, and this will certainly affect chips, especially with the spread of local LLMs. At the same time, this is a good sign of increased demand for resource optimization and AI products, a niche for startups, so to speak.

A friend of mine told me a funny story about this. Even before the AI hype, he made a generator of unique texts, posted it on his domain, and forgot about it. The script generates a large number of random words and sentences each time it is accessed. So, the other day, he decided to look at the log statistics.

ClaudeBot/1.0 turned out to be the leader in terms of requests: more than 400,000 requests per day, downloading 46 GB of text in two days. Just garbage text. And it continues to do so. And how much of this garbage is there on the internet? Imagine how much energy and other resources, including venture capital, are being wasted.

5. The AI generation is born

For children born in the last five years, neural networks will be a basic technology. Their experience of interacting with machines will be radically different from previous generations, and the opportunities this will open up will be enormous. Over the past 20 years, the Internet has radically changed both the level of information accessibility and the speed of that access. Not only consumption but also creation is becoming much more accessible. Skills that used to take years to master are now being taken over by machines, and children will be able to grow up building their world with greater freedom for new ideas.

6. New risks and new professions

People often use products with neural networks as a black box. For many, it is almost indistinguishable from magic. In addition, we are lazy and happy to rely on something that does the work for us. This habit develops very quickly and willingly.

All of this creates new risks- security, errors, and misinformation. New vectors of attack are emerging for the introduction of malicious instructions to AI agents. Vibecoders create applications that turn out to be full of bugs if made without proper control. Not to mention the temptation to use LLM chats for work documentation and communication. Dozens of times in my work, I have seen people send absolute generative nonsense to Jira, by email, or in chats.

In addition, AI systems themselves have become multi-layered - prompts, agents, intermediate data processing services, specialized neural networks for specific tasks, model adaptation, and so on.

This is already a big world where new professions of various levels are emerging and will continue to emerge. Starting with ML Ops, Data Engineering, and other purely technical specializations, continuing with new areas of security, working with specialized models in various subject areas, managing AI agents, retraining, using assistants, and writing high-quality prompts.

Just to maintain the global AI ecosystem, many new skills and specialists will be needed. But I also have a positive view of the “old” professions. There is still a shortage of labor in the world, from drivers to doctors. And entry into many non-critical professions will be reduced due to AI systems. More people will be able to work in jobs that were previously unavailable to them due to lack of education and high costs. An analogy can be drawn between a car with a manual transmission and a car with autopilot. We still drive the car, but it requires much less knowledge and skill.

7. Local models and market fragmentation

Everyone is talking about OpenAI, Claude, Grok, Gemini, and Deepseek right now. However, there are a sufficient number of open source models, such as Llama, Qwen, and others. On powerful hardware like Nvidia DGX Spark, you can get results that are no worse than those produced by Google's Nano Banana.

All this allows you to build a private infrastructure for business or even for individual needs. Given the need for information security and privacy, flexibility for specialized tasks, and cost optimization, both hardware and local models will continue to evolve. This will lead to market fragmentation and new opportunities for niche products and specialists.

8. Global adaptation

Billions of people in dozens of countries are far from the AI hype; for them, these technologies are simply inaccessible. In many places, even simple digital services have only recently appeared or are just beginning to appear. However, where there is virtually nothing, especially in developing countries, everything new will be built immediately using AI - automation of production, services, the consumer segment, and so on. And adaptation to real life will probably be more organic and rapid. For example, as was the case with banking services in Russia, when everything has to be built from scratch, the use of current technologies becomes a self-evident solution.

Perhaps in 10 years, we will meet tourists from African countries who are surprised by primitive mobile applications in which you have to tap buttons with your fingers to order a taxi.