AI will replace you!”
Have you heard that? I hear it often, and I wanted to figure this out and understand for myself how software development will live in the conditions of universal AI, when just about anyone can use AI to generate a website, a bot, or use a neural network to generate applications. And it seems that, indeed, programmers are no longer needed, but intuitively we understand that it is not that simple.
Let’s figure out who a developer is and how they differ from a programmer, an architect, and an agricultural engineer.
But first, let’s conduct a thought experiment.
Suppose you need to move into one of two skyscrapers on the 99th floor. Both skyscrapers have 100 floors, foundations, the same amount of apartment and common-area space, and each has a built-in fire safety system, but in neither case has it ever been activated. But there are also differences: the first building passed a full inspection by the relevant state fire, technical, and consumer-safety authorities, and it was designed and built by people according to all modern standards. The second building was built by AI without the involvement of human specialists. The relevant state fire, technical, and consumer-safety authorities did NOT participate in putting the second building into operation.
Which building would you prefer for the next 5–10 years?
For some reason, it seems to me that the first one.
We can conduct a similar experiment with doctors. If you eat the wrong mushroom, who would you go to for treatment? A human or AI? For some reason, it seems to me that everyone would go to a human doctor for the final treatment. And that is understandable. We do not yet trust AI that much, especially when it comes to things like health.
And the reason is not that AI is bad. And it is not even that the words of a human doctor can soothe our emotional traumas. The reason is the same reason why people choose the creative work of other people.
AI music has surpassed 50% of new music uploads
The study “Human Bias in the Face of AI: Examining Human Judgment Against Text Labeled as AI Generated” and the article “AI music has surpassed 50% of new music uploads for the first time” by Deezer clearly point to this. The latter gives some interesting numbers: more than 50% — around 90,000 — of all tracks uploaded to the platform every day are created with AI, while the share of streams of such tracks is only 1–3%. Makes you think, doesn’t it?
Let’s nevertheless get to the main question. What is an architect, what is an agricultural engineer, and how is an architect or agricultural engineer different from a developer? All these specializations have one thing in common: specialized knowledge and responsibility taken upon oneself.
A building architect is responsible for ensuring that a design complies with applicable standards and requirements. A bridge architect is responsible for ensuring that loads, operating conditions, and safety requirements are taken into account in the design. A software architect is responsible for architectural decisions that allow a system to withstand loads, scale, and evolve. A developer is responsible for ensuring that the software they create can be maintained and used for its intended purpose.
And a programmer writes code. Of course, they can solve algorithmic problems, pass interviews, and follow instructions… But for some reason, it seems to me — and I am almost certain about this — that neural networks do these things better and faster. And therefore, the task of every programmer is to become a developer.
Agricultural engineer.
I did not choose this profession by accident. We can look at the evolution of this field roughly, and we can already see new synergies between AI and agriculture.
Until the 19th century, fields were cultivated with plows using people and animals. Later, steam engines and internal combustion engines appeared, and at the beginning of the 20th century, the first motor cultivators appeared. By the first half of the 20th century, tractors began to spread widely, and by the end of the 20th century, tractors and combines had become an integral part of large-scale agriculture around the world.
Then mechanization began to transition into automation of technological processes and production: electronic controllers, sensors, and automatic control of individual operations appeared.
Later, satellite data and remote sensing came into agriculture. Today, there are already projects where computer vision and AI are used to identify and map weeds, estimate their density and biomass, and detect plant diseases.
And every stage was accompanied by the elimination of a large amount of simple work and an increase in the importance of more specialized knowledge.
The same thing is happening in IT.
Within IT, every developer is first and foremost an engineer, and only then a programmer. Programming has constantly been getting rid of low-level routine work. We moved from machine instructions and assembly language to high-level languages, libraries, frameworks, and IDEs. Each such stage removed part of the routine and allowed developers to operate at a higher level of abstraction and develop more complex software.
Generative AI is taking the next step — now the routine work of describing functions can be handed over to the machine itself, while we focus on architecture and software design.
That is why AI implementation in business processes and AI implementation in processes in general are becoming less a question of code generation and more a question of properly designing these processes.
The same applies to process automation. If previously a developer automated an individual operation using code, now they can design an entire system in which some of the operations are performed by AI.
A few words about vibe coding.
Most people will not build a multi-story house themselves without the help of specialists. I believe that each of us can build one or two floors ourselves, but building three or four without the necessary knowledge is already dangerous. Anyone can run a wire to move an electrical outlet, but organizing all the wiring in a house so that it can handle the loads requires knowledge. And we all understand this perfectly well.
We know what to do when we have a cold, but if something more serious happens, we go to a specialist. We know the approaches of juniors and some middle-level developers very well. We have experienced all the pain of incorrectly chosen architecture ourselves.
The euphoria will pass, but hundreds of thousands of projects killed by vibe coding will remain, and they will require specialized knowledge.
That is why implementing process automation with AI is not simply about connecting a neural network. You need to understand which processes should be automated, where a human is needed, where a model is sufficient, and where full engineering control is required.
I am not an enemy of vibe coding. It is a wonderful tool that requires proper handling and precisely those specialized skills in architecture, development, and even programming.
Conclusion
And when it comes to complex systems, implementing AI agents in business processes requires an even higher level of understanding: an agent must not simply generate an answer, but operate within specific business constraints, use tools, access data, and handle errors correctly.
Otherwise, it will be like with music: 97% of people versus 3% AI. AI is already capable of producing a gigantic amount of content. But the Deezer story shows an interesting paradox: production has become cheap. Something else has become expensive — creating something that someone actually needs.
Perhaps this is exactly what is happening with programming right now.
Code is becoming cheap. Engineering is not.
Top comments (0)