The first victim of the AI ​​cyclone: ​​the world of software

by Luca Indemini
Fonte: La Repubblica

The ability to write code without knowing programming languages ​​has revolutionized the industry. But there is a crucial point to keep in mind: artificial intelligence does not reduce complexity, it shifts it.

What is happening in the software world with the arrival of generative AI—and in particular with “agentic” AI—is not a simple technological evolution, but a profound redefinition of economic value, professional roles, and the architectures on which products are built. The transformation significantly impacts the required skills and the organization of software development. According to Google’s Dora State of AI-assisted Software Development 2025 report, approximately 90% of developers use AI in their work, and tools like GitHub Copilot have become de facto standards.

The signs are already evident even at the macro level. The sharp decline that hit companies like Atlassian, ServiceNow, Salesforce, and Microsoft in early 2026 reflects a dual tension: on the one hand, the fear that AI could squeeze the margins of traditional software, and on the other, the enormous capital requirements of algorithmic infrastructure, with investments expected to exceed $700 billion by 2026.

The question that emerges from these premises is: what is really valuable in software, in a world where code is increasingly a commodity?

“The value no longer lies in the interface designed for human interaction,” emphasizes Eric Clark, CEO of Manhattan Associates. “With the arrival of AI agents, the value lies in the data model, the APIs, and the semantics of workflows.” In this context, AI does not erode value, it amplifies it. Because efficiency no longer comes from how intuitive an interface is, but from how accessible, modular, and interoperable a system is.

This transformation is also directly reflected in the work of developers. If until yesterday, programming primarily meant writing code, today the required skills are changing: “It will be less and less a programmer and more and more a designer,” explains Elena Baralis, Pro-Rector of the Polytechnic University of Turin and professor in the Department of Control and Computer Science (Dauin). “Prompt engineering” is becoming a key skill.

Or to put it in Clark’s words: “At Manhattan Associates, humans define what the system should do; machines take care of how. Human judgment moves higher: into architecture, domain modeling, governance.”

The use of artificial intelligence drastically reduces the cost of producing code, especially standard and repetitive code. As a result, the most exposed roles are those related to mechanical coding and entry-level software, while the value of those who can design complex systems, manage architectures, ensure security, and govern the software lifecycle is growing.

This is confirmed by an analysis by Data Masters, according to which developers who master generative AI and advanced architectures earn paychecks up to 15 percent higher than traditional profiles.

The problem?

These figures are still rare, compared to the demand.

Programmers are therefore required to have more “abstract” and higher-level skills. Non-university training courses will likely be the ones most affected if they fail to transform in time. Universities will increasingly be tasked with developing critical thinking skills and providing solid conceptual foundations, regardless of the tools used. “At the Polytechnic, a few years ago, sensing the revolution coming, we decided to experiment with a course that combined Large Language Models and software engineering,” explains Baralis. “It’s currently a master’s degree program, but the goal is to extend similar courses to the undergraduate program, to immediately provide our students with the necessary foundation.”

Perché se la scrittura è sempre più appannaggio delle macchine, «il problema della responsabilità resta umano – sottolinea Riccardo Coppola, titolare del corso Large Language Model per l’ingegneria del software al Politecnico di Torino –. Così, se da un lato il ruolo del developer come lo abbiamo conosciuto fino a oggi è destinato a ridursi significativamente, diventerà sempre più centrale il lavoro di verifica: code review, validazione, governance degli strumenti IA. Senza dimenticare le implicazioni etiche e normative: gestione dei dati sensibili, compatibilità con il Gdpr, controllo dei bias presenti nei modelli».

A crucial point – often overlooked – is that AI doesn’t reduce complexity: it shifts it. The more code is produced automatically, the more critical it becomes to understand whether it works, whether it scales, whether it is secure. E qui la sfida per il mondo accademico si fa ancora più pressante. Senior programmers once trained primarily on the job; if AI is destined to drastically reduce the number of junior positions, the question is: how will tomorrow’s seniors be trained? Those who should manage and control AI? Coppola e Baralis concordano su un punto: «Il coding non deve sparire dagli atenei. A scuola continuiamo a imparare le divisioni e le moltiplicazioni anche se ci sono calcolatrici che le fanno più rapidamente, servono a formare la nostra capacità razionale. Quindi, usiamo l’IA come fosse una calcolatrice molto potente, ma analizziamo i calcoli, continuiamo a verificare i risultati».

Eric Clark agrees on this point: “If you use AI to take low-level work away from people, you’re not just saving time, you’re eliminating scale. Many of today’s senior engineers grew up writing trivial code, debugging complex errors. AI must be seen as a support element, a tutor, not as an oracle. This way, tomorrow’s seniors will be those who have learned to use AI, question it, and test it.”

E qui il tema si ricollega anche al futuro del Software as a Service. The common belief is that by replacing programmers with machines, AI will reduce the cost of coding to zero, causing SaaS margins to collapse. But this is a misleading idea: while AI reduces development costs, it also increases execution costs, making platforms capable of managing and optimizing these aspects even more important. “The economy is changing and the premium for platforms that can manage this complexity on behalf of customers is rising,” Clark emphasizes. “Software as ‘code’ is becoming a commodity; service is the determining factor.”

The process is already underway: the center of gravity is shifting from code to architecture, from interface to data, from manual execution to automated orchestration.

Agentic AI does not eliminate software, but redraws its value hierarchy. The companies that can build solid, integrable, and manageable systems—and the professionals capable of designing them—will be the ones who will truly benefit from this transition.

The question, then, is not whether AI will replace programmers or traditional software. It’s that the definition of both is changing. In this context, a “majestic designer” will be increasingly necessary, as Elena Baralis defined it: “The real challenge is not technological, but cultural and educational.”