Opinion piece / AI boosts efficiency, but does not understand

There has been a great deal of heated debate about the effects of AI on working life and in the IT sector, especially on software development. However, most of the discussion and writing remains at an optimistic, high level: it talks about productivity, efficiency and opportunities. Less attention is paid to the moments when AI is not enough or when it misleads its user.

AI has the ability to significantly streamline individual tasks for a software developer. Writing code is a good example of this. With the right prompts and access to the right data, AI can, for example, write code fairly competently, but it is by no means a true software engineer. Software projects are first and foremost thought work rather than “just coding”, so high-quality, maintainable software cannot be built by combining individual code snippets produced by AI. AI, however, cannot grasp context and the big picture.

How could it, when describing the business environment to AI in all its complexity is practically impossible? And it is precisely these complexities that set the requirements for software engineering.

This is reflected, for example, in the fact that AI often provides the first factually correct answer to a given prompt – but it is not able to provide the best, most maintainable or most understandable answer. A programmer who has worked with AI less or is less experienced may accept the machine's answer without much question, and the real consequences of the choice may only emerge months later. Even extensive use of AI as an aid does not remove the need for a person with a broad perspective on both technical and non-technical matters. In addition, a software developer must be able to think critically in order to recognise when AI is leading them in the right direction and when it is leading them in the wrong one.

Learning cannot be outsourced

Blind trust in AI is especially worrying from the perspective of learning. If people studying programming become accustomed, already in the early stages of their studies, to AI always handling certain tasks, they may not learn to distinguish a sensible solution from a nonsensical one. At the same time, the ability to challenge prevailing solutions disappears. Learning is about understanding, constructive doubt and weighing up alternatives. If the thinking work required for learning and the actual doing is outsourced to AI, all that remains is mechanical performance, and genuine learning does not take place. A question also arises: if the thinking work is outsourced, who checks the choices made by AI? And if a human has to check and confirm AI's work anyway, is AI ultimately of any benefit?

Many technical professionals have a dual attitude towards AI, as they see the opportunities, but also AI's shortcomings. In public debate, critical voices often get pushed aside when the focus is on inspiring stories and the growth figures of AI companies. AI must also be approached critically.

Regardless of viewpoint, the age of AI can no longer be stopped. The change is irreversible, and it is happening right now. The time when every line of code was written by a human is over. But AI does not yet today – and perhaps not in the near future or perhaps ever – replace the real software developer and the work of thinking.

We cannot know where AI will eventually take us. In five years, the situation with AI may look completely different. But we can decide how we respond to it. If we want to build smart systems and sustainable working lives, the discussions we have must include dimensions other than efficiency.

Arttu Ylärakkola

Founding partner, software developer, Rakettitiede

This article was published in Tivi in June 2025.