The amount of text, images, and videos produced by large language models (LLMs) could well be on par with the enthusiasm generated by this new technology. Just think of the volume of investments made in recent years, the proliferation of companies and services, and the public attention surrounding this artificial intelligence tool. It feels like we are caught in a race moving at unprecedented speed. And yet, in the recent past, we may have experienced similar dynamics. This is the view of Tiziano Leidi, Director of the Institute of Information Systems and Networking (ISIN) at SUPSI.
“However fast they may seem, I believe the timelines are comparable to those observed with the advent of the internet or the personal computer. In the research field, natural language processing models were already being developed, for example those of the BERT family (Bidirectional Encoder Representations from Transformers, published by Google in 2018). For the general public, of course, everything changed with ChatGPT. A model was made available with sufficient capabilities to enable conversational interaction. Only a few years have passed since its release, but the reality is that there was also a long ‘before,’ and above all that there is now an entire future ahead. If we look at what tools based on LLMs allow us to do and where they fall short, we realize that we are still at the dawn of a technological evolution that will have a great deal to offer in the coming years.”
Tiziano Leidi, Director of the Institute of Information Systems and Networking at SUPSI
Drawing further comparisons: while we are conducting this interview (it is late November 2025, at the time when major high-tech companies are presenting their quarterly results), more than one analyst has begun to speak of a speculative bubble around AI. Some even compare it to the dot-com bubble that burst in 2000. While it is true that a bubble can only be identified once it bursts, it is also true that the volume of investments brings to mind what happened more than twenty-five years ago.
“I wouldn’t be so quick to compare what is happening with AI to the dot-com bubble. There are probably some similarities, but also many differences. In general, there is a lot of hype around AI, with investments that may be excessive relative to the current state of the art, much like what happened in the late 1990s. At that time, there was heavy speculation on small startups that had no real product and were unable to generate profits. Today, speculation is concentrated on large companies in the sector that are investing billions in a technology with enormous potential, supported by high-performance hardware infrastructure. The foundation, in short, is more solid and resilient. That said, it must be acknowledged - and this is a fairly widespread view -that compared to the real needs of those who use this technology, the capital being moved in the competition among major players is exorbitant. I believe that at some point we will have to go through a corrective phase: whether this will involve the bursting of a bubble or a gradual return to a more sustainable equilibrium remains an open question.”
Indeed, there is little doubt that sooner or later all this excitement will fade, and the novelty effect will be followed by a new phase in which LLMs will become a normal, perhaps even taken-for-granted, presence in our lives.
“We are in the middle of the hype phase, and this is usually followed by a phase of disillusionment. The question is how deep this user disillusionment will be - in other words, what will happen when we realize that many of our expectations cannot be met with current technologies; to what extent will we be able to settle for what AI can actually do? The constant stream of announcements is exacerbating both excessive enthusiasm and catastrophic thinking. I believe there are both opportunities and risks. We need to find a good middle ground, leveraging the potential improvements these technologies could bring to our lives.”
Speaking of the announced innovations: there are many, they are frequent, yet the impression is that after bringing these technologies to the general public, there have been no developments of comparable magnitude.
“The radical innovation arrived with the launch of ChatGPT, which surprised everyone with its conversational capabilities. From there began what in English is called ‘commoditization,’ meaning that this type of technology has now become standardized and uniform. There have been constant developments in its application: from text generation, to images and videos, and now it is moving into 3D. Alongside commoditization, architectures have begun to emerge -such as agent-based systems -in which multiple language models communicate with each other to produce results in increasingly concrete applications. In computer science in particular, tools are emerging that enable synergy between developers and AI, accelerating developer productivity.”
Is it therefore conceivable that the code-writing component - which is often the first image associated with developers - will gradually become less dominant in your work?
“Today, developers who are valued are those capable of conceiving a product, engineering it, and defining its requirements and functional specifications. Through interaction with AI, the process of building the artifact begins, using tools that until recently were not available to us and that can automate a whole range of operations within the complex process of software development. In my view, the true nature of this technology is to support development and increase the quality of the final product. Moreover, AI facilitates error detection in code, cybersecurity management, and the automation of operations in complex environments. So yes, the developer who merely programs based on given specifications is a figure that, over time, could disappear.”
This does not mean that there will no longer be a need for professionals.
“Quite the opposite. Think about the apps on our smartphones - over the next few years we will see them transform. The IT world is in constant evolution and is growing in complexity. Consequently, the need for professionals will continue to grow. Developers will need to update their skills to keep pace with technological evolution. LLMs today are available to everyone and can contribute to product innovation. Many have recognized this opportunity and have launched new startups or are innovating processes within their companies. Those who fail to keep up risk being left behind. This is part of the very nature of the engineering profession. At SUPSI, we are committed to training new professionals with knowledge aligned to the state of the art and with the mindset necessary to best interpret technological evolution.”
Also because, powerful as they are, today’s LLMs are not able to generate ready-to-use medium-to-large software systems - if they ever will be.
“I am amazed by the ability of these language models to produce solutions that at least move in the desired direction. But we are talking about solutions of relatively small scale. Human beings play a central role in reviewing and providing feedback on the output produced by LLMs. In highly complex systems, however, it is difficult to obtain robust results without continuous interaction between humans and machines. Despite what many companies in the sector claim - which I consider largely marketing operations - I seriously doubt that we will soon reach a point where we can ask AI to develop software to manage complex tasks and receive a finished product in return. To draw a comparison: it is conceivable that in the not-too-distant future there will be robots capable of building a bridge, but there will still be a need for human beings to define how these robots should behave and to ensure that they actually complete the construction correctly. The connection between these two worlds will be the winning factor.”
Looking back, how much has your work changed with LLMs?
“In November 2022, when ChatGPT was released, some researchers at the Institute knocked on my door to talk about this new tool. Three years have passed, and I can say that it has significantly changed our work. In addition to helping us with translation and text revision, it has allowed us to save a tremendous amount of time that we previously spent browsing forums in search of solutions to problems we were facing. Setting aside a few hallucinations or minor errors, the language model is an extremely effective advanced research tool and is certainly faster than spending hours in a forum searching for an answer. With the arrival of agent-based architectures, which are very topical today, we have begun to use these tools mainly for the initial prototyping of application components. Today, we delegate certain fairly mechanical tasks to the language model, which generates hundreds of lines of already organized code that we can then refine if necessary.”
“From this perspective, we expect LLMs to become increasingly capable and, on demand, to materialize portions of programs that still need to be integrated and supervised. The software development process will be understood as a dialogue between humans and machines. Tools already exist that enable spec-driven development (SDD), a specification-based approach in which the developer describes what they expect the program to do in its various parts and then, step by step, proceeds incrementally with the language model to build it.”
In summary, it would not be wrong to say that the programming world is going through a major period of change. Is this already being felt in Italian-speaking Switzerland?
“The IT sector in Ticino has always been significant and has grown over the years. Alongside medium-sized companies, smaller ones are now emerging as well, thanks in part to the advent of AI solutions. There is an effort to seize this opportunity to offer new solutions. In my opinion, the community is extremely active—we see this at SUPSI too, in the demand for professionals and in the growing number of students. Young people who see an important professional opportunity in this field. What has not been lost over time is passion: we still find many young people driven by an interest in software engineering who study to become highly qualified professionals.”
A moment from the inauguration of last year’s Voxxed Days Ticino
For the community, an important opportunity to come together is Voxxed Days, which this year will take place on Friday, February 6. Will LLMs be the main focus?
“The themes of our conversation will also be reflected at Voxxed Days. We will talk a great deal about new development tools built on large language models. We will maintain that link between past and future, both in programming languages and in the culture of the IT world itself. We have received a very large number of high-quality submissions and have decided to increase the number of talks. Beyond the content, we must not forget that Voxxed Days are also an opportunity to meet, network, and connect - and that aspect will certainly not be missing.”