Short answer
AI can automate routine programming tasks and accelerate development. However, creating complex software products still requires experienced engineers who can design architecture, understand business logic and oversee the quality of the result.
Developers who have mastered AI tools can significantly increase their productivity. Beginners, meanwhile, should use AI not only to generate code but also to learn, analyse solutions and develop their own engineering skills.
For a business, what matters most is a team that can use AI effectively while remaining responsible for the quality, security and reliability of the software product.
Preface.
I was prompted to write this article by numerous interviews with Demis Hassabis (one of the most influential scientists in the AI field) and by a social question that has emerged about whether programmers will be needed in the future and what their employment will look like in an era of active AI adoption.
Let’s begin.
There’s a widespread opinion right now that AI will replace programmers, millions of people will lose their jobs, and programmers will become unnecessary altogether. I’m very skeptical about this claim. At the same time, our team actively uses almost all available tools provided by LLMs (large language models — for simplicity, I’ll just call them “AI”), and based on the expertise we’ve built up, I’m ready to draw some conclusions that I want to share with you.
AGI vs AI.
First of all, people often interpret AI incorrectly, confusing it with AGI (artificial general intelligence), and they attribute to the already familiar LLMs (AI) characteristics these systems are not capable of. At the same time, AGI does not exist at the moment. In the theoretical concept, AGI is meant to be a self-learning system that should perfectly understand cause-and-effect relationships and act according to tasks and changing conditions, while controlling itself. AI, on the other hand, is a technology based on LLMs (large data models trained by humans) that is designed to be a tool for automation and for solving tasks that previously only humans could perform. In other words: AGI is an independent entity with human-level intelligence, while AI is a helper technology.
We’ve sorted out the theoretical part regarding AGI and AI, but there’s one more question that needs to be answered so we can fully put the formula together and move on — we need to define who a programmer is and what this profession actually is.
Who programmers are.
If we look at a programmer as someone who simply writes code, that’s the main misconception that creates confusion. A programmer is not just a “function” like a button — “on/off” or “write code.” First and foremost, a programmer is an engineer who defines the architecture and the fundamental points of a project, with future scalability in mind and with consideration for the set of technologies that will (or might) be used in the future. Any growing project constantly acquires new sub-architecture as it evolves and requires taking into account both the existing foundation and the implementation of new features under all requirements. That means the architectural task remains throughout the entire development cycle.
I’ll allow myself to say that a programmer “as a function” plays this role only when fixing an obvious bug (when you can clearly see it on the screen) by correcting a few lines of code. But there are also errors that arise from calculations and are related to business logic, where you don’t just fix a couple of characters to eliminate a syntax error — you need to communicate with the client to clarify all criteria and calculation formulas (for example, in financial modules).
Now I’ll give a more understandable example. You’re building a house. First, a plan is created, and only after that do specialists come in to lay electrical wiring, build water supply systems, and perform other types of work based on the described task. Without a plan, it would be chaos — everyone would do the work however they wanted, and the end result would be exactly like the task itself — nonexistent. And when something needs to be modified in the communications where there is direct access, the task still isn’t solved mindlessly, because you need to consider power consumption, safety, operating conditions, and other factors before proposing a solution. Even something as simple as extending a cable is handled by a competent electrician with regard to the wire gauge, material, and other characteristics. In other words, there’s always an analytical phase before simply doing something.
In our example: the house itself (its frame) is the base architecture; the interior work requires a sub-architectural process; any other work with different parts of the house also requires a sub-architectural process. To put it even more simply: hammering a nail is a function that requires minimal effort; all the calculations (where, how, with what force, and most importantly — for what purpose) are the preliminary planning that requires analysis.
Back to programmers and AI.
We’ve probably spent quite a bit of time on the theoretical part (and I hope I managed to convey that any process consists of planning and implementation), and now I’m ready to share my conclusions about what awaits programmers. And here I have great news for true engineers devoted to their craft, good news for beginners, and very bad news for those who deny the revolution.
I’ll start with the unpleasant part, and I really hope that the number of people in this category will keep shrinking.
For all programmers.
If a programmer writes code manually and still uses traditional tools to search for solutions, that engineer has every chance of losing their job — and not because it’s bad, no. It’s simply because today this can be compared to about 10% efficiency of all the possibilities that LLMs can offer (ChatGPT, Gemini, and others). It’s like inventing the wheel and still carrying sacks on your shoulders over long distances, denying revolution and progress. At the same time, I’m not advocating a process where an employer or client squeezes maximum efficiency out of you to take all your resources for pennies. The point is that for me, as a programmer with more than 20 years of experience, there are obvious things today: the era of searching for answers and solutions as we knew it is over. And that means one thing — we can grow faster and find solutions to complex problems faster, and those who don’t use this simply lag behind their colleagues.
For beginners.
Here, of course, things are not so simple or straightforward, but I see more positives in the fact that colleagues with little experience can learn faster. By asking an LLM questions with a description of your task, a beginner will get code examples where the AI tries to assemble the needed solution, and learning from examples is always easier — especially when the example is built specifically for your task description. Of course, sometimes formulating the question can take longer than writing the code itself, but the key point here is learning from examples: you can see new and interesting approaches you didn’t know about before. Over time, as a beginner starts solving more complex tasks, they’ll formulate questions to AI more precisely, understand its answers faster, and increase their efficiency.
Can AI harm beginners? Absolutely yes — if you just take the code, don’t verify the result, and don’t learn, it will have negative consequences both for the specialist and for the company they work for or whose tasks they solve. But I still choose to look at this positively, because LLMs can provide significant progress in the learning process, and I hope this is exactly the scenario that awaits all my readers.
For true professionals.
And now the most pleasant part! Programmers who have grown and honed their experience over the years have received an incredibly powerful assistant in the form of LLMs. They now have time not only to optimize the routine they’ve long been tired of, but also a great opportunity to multiply their efficiency. They already know how to build architecture, how to write good code, and most importantly — they understand growth and product scaling. What they sometimes lack is speeding up the writing of repetitive pieces of code and accelerating the search for solutions to tasks related to integrating external systems (for example: WABA, Stripe, and other APIs). That is, when they need to figure out something new or something they’ve forgotten, they can now do it many times faster and with far less personal effort. Such a professional turns into a true maestro, orchestrating the assembly of complex systems. They “use the wheel” instead of denying its existence, they bring more value per unit of time and with far more enjoyment than they could before LLMs appeared. A new era will begin faster for these people, because they’ll immediately use the tool as intended.
For experienced programmers, an incredible opportunity has arrived — to very quickly become part of a new, irreversible process and master a new branch of our profession, which I would boldly call “development using the skills of an LLM operator-engineer,” and to use this new tool for the benefit of themselves and their employers or clients.
For clients and employers.
If I were in the position of people or companies ordering programming services, I would start asking whether the team building their products uses AI tools — but without fanaticism and with an understanding of the essence of the issue (that’s exactly why I’m writing this article). It’s like simply asking your construction contractor whether they have all the tools needed for the job, or whether you’ll have to buy them yourself. In our example, that would mean reduced team efficiency and a slower development process without any objective reason.
For those who think they can do without a programmer.
Perhaps one of the most interesting and sensitive questions, which is actually the root of this entire article. LLMs really do make it possible to enter a prompt like “build me a website page that contains …” and with 99% probability you’ll get a result that makes you happy, because you did it yourself. If for a business it’s enough to get such a result, deploy it on hosting, and consider the job done — all that’s left is to applaud everyone involved, because the task has been solved.
When it comes to building complex systems that require data management, payment and transaction control, user logic, and so on, AI alone is not enough, because you’ll end up with code that you still need to do something with. And if after getting this code you know what to do with it, apply your knowledge to integrate it into a project, and everything works — that means only one thing: you are a programmer (at least a beginner). If, however, someone without programming knowledge decides to take their project’s code and, through special editors (IDEs), give AI commands to refine the product, the result will be roughly the same as taking apart a modern car’s assembly, bolting on new parts, and expecting everything to work by itself — with new and improved characteristics.
Even in the most high-precision manufacturing, there are special machines that verify results — in other words, there is an “observer” that evaluates quality, and there are people who make decisions and manage the process. In this regard, AI has not caused a revolution, and its creators never had such a goal. The task of LLMs is to be a tool and an assistant.
One modern misconception that has arisen due to aggressive marketing in the AI industry is the attempt to apply AI technology, which can reason beautifully, to processes that require extremely precise decisions and actions.
In this chapter, my verdict is:
1. For simple solutions — simple tools (they already exist: cloud website builders, chatbots, and other micro-technologies, if they don’t require customization for your business logic);
2. For complex solutions — engineer-programmers with AI skills;
3. For tests and satisfying curiosity — any means, including trying to figure things out and “coding” with AI on your own.
For those who keep claiming that AI will replace developers.
I’ve tried and spent quite a lot of effort to systematize my knowledge, provide examples, and offer balanced arguments. At the same time, I understand that everyone has the right to their own conclusions. Looking at the picture through the eyes and mind of an engineer, I can’t imagine how “a hammer can replace the specialist who uses it.” Make the job easier — yes. Replace — no, due to its nature and purpose.
Summary
I’m very inspired by the changes that have arrived. I have a feeling that this is exactly what was supposed to happen — at least in the programming sector. Everything has become very complex. For example: just 10 years ago, we had minimal requirements for the visual part of a simple registration form and its logic, and it could be built in 10–15 minutes. Today, building a user registration form can take half a day if you implement multiple validations at the input stage, account for multilingual support, animation effects on submission and validation, integration with third-party registration services via Google and Apple, SMS verification codes, and so on. Something like this had to happen so that part of the routine programming work would fade into the background. The programming community and private companies addressed these issues by creating special libraries (that’s what we call them) to simplify repetitive tasks, but over time there became too many of them, while the requirements for modern products only kept growing.
Now everyone has received a unique tool in the form of LLMs (AI), with which developers save time on routine tasks, and clients and their projects keep up with modern requirements. In my view, this maintains a certain balance, because we simply can’t spend dozens or even hundreds of hours on a single, not-so-critical system component. For programmers, that’s boring; for clients, it’s expensive. It’s better to focus on what really matters and on processes that help businesses grow, because that’s exactly why we (programmers) are here.
P.S. I’m eagerly awaiting AGI, which I mentioned at the very beginning of the article. If humanity manages to create something like that, it will be the starting point of a global revolution across all service and manufacturing industries.
Thank you for reading my article.
Respectfully,
Co-founder of DevCats, Ihor Tykhonov
Updated: