CTRL/SHIFT - 15 06 2026 PARTHENOPE STAGE - ALESSIO POMARO - TRASFORMAZIONE ESPONENZIALE

We are in the era where artificial intelligence is capable of generating research, analyzing data, and automating all activities. It is even said that the first single billionaire will be able to exist thanks to AI, which could have 200 agents, each dealing with marketing, budget, advertising and much more and together for this. So I immediately give the floor to Alessio who is very happy to accept the invitation. Thank you. Thank you.

Applause. This is good, you can feel it. I imagine that each of us has made popcorn at least once. So how does it work? We put the corn in a hot pan and initially nothing happens, it seems like a very slow process.

After a few minutes, however, we hear the first grain pop, then immediately the second and in the moments immediately following all the others pop, practically all at once. Well, I believe that as far as the development and integration of AI are concerned, we are at the point where we have heard the first grain explode. To put it outside of the metaphor, I believe we are at the beginning of an exponential transformation— not a rapid one, but exponential in the true sense of the word. And we already have the first signs of all this. A small but important example comes from Google Deep Mind's Alpha Evolve project, an AI-based system capable of autonomously optimizing programming codes.

This may seem like a trivial operation today, but Google has already integrated it into its ecosystem, achieving truly interesting results, such as resource optimization, the discovery of new algorithms, a boost in autonomous scientific research, and hardware optimization. Yes, AI is already helping create better chips, which will then train and run better AI, which will then help create even better chips. This is a typical example of one-moment dynamics, eh, one-moment exponential dynamics. We're already in a situation where AI is creating algorithmic solutions that are often counterintuitive even for human experts. However, they are opening up new scenarios—new scenarios, we might say, for scientific discovery, for AI architecture, for industrial optimization.

Not only does Stanford's Rebard AI Index show how, for example, AI is already outperforming humans in several benchmarks, but it also shows us something else: measuring the performance of these systems is becoming increasingly complex. On the other hand, we have the AI ​​Security Institute UK which shows us how the complexity of the tasks that these systems can handle autonomously is now doubling every 4 and a half months. Well, let's take a closer look at what this growth in the models we're using means, and we'll do so by exploring the visual world because it's the most immediate to understand. For example, who remembers the first images that were generated through an AI model? We have come across monstrosities like this, salmon swimming in the river depicted in this way.

Oh, I know, it's funny. But that happened two and a half years ago, because today those salmon are represented like this. And what about the video generation, where are we at? Someone has already shown it before. I'll show you an example too because we're already here with the video generation.

Now the of my never is tr and to you my to Gotevole, but certainly the real impact will not come from increasingly qualitative images and videos, but will mainly come from the changing way of working of teams and will be increasingly attributable to the processes that change and the companies that change. So what does it mean to integrate AI into business workflows today ? Today, in my opinion, integrating AI means moving from adoption to transformation. And this has two components which are thinking and AI first and application layer. Let's go and see what each of these two elements means and let's start thinking AI first.

Satianadella, CEO of Microsoft says: "Every company will become an AI company, completely transforming the way they work." In fact, thinking AI first means not applying AI to optimize consolidated processes, but it means reconstructing those processes with AI at the center, that is, moving from asking how can I automate this report, to instead asking what decisions will this report allow me to make and how do we get there? In my opinion, this is achieved through two ingredients: actively cultivating curiosity and experimentation at all levels within the company, with company leadership encouraging this type of process—that is, encouraging the questioning of established processes to see if they can be improved and revolutionized through AI. The other ingredient is this: cross-functional collaboration with, for example, the company's IT team, which is responsible for consolidating and aligning with the business what is being achieved in the testing phase within the various teams. I know what you're thinking: that these are all theoretical concepts and difficult to apply.

And so I'll tell you how this way of thinking happens in our organization, namely In Search on Media Group. In research on the research and development and consulting teams of which I am a part, they create proofs of concept, they create prototypes to improve the workflows for the processes in the company. And how do they make these prototypes? using, for example, AI-based agents that assist in the development of code and applications. For example, Open AI's Codex is this, but there's Cloud Code which I imagine you're all familiar with.

The development team is responsible for creating guidelines that support the teams in their experimentation phase, for example by creating specific skills that help them in this phase and obviously also takes care of consolidating what is developed in research and development. Management encourages this type of activity, so much so that these proofs of concept developed within the teams then become services to be made available to customers and everything is aligned with the business. It is no coincidence that these services are merged into a SAS platform that we then make available to customers and also to other users who wish to subscribe to this system. So let's see how from a proof of concept within the teams it becomes a service and then it becomes a product. Keep this in mind because later I'll show you a little preview of this product.

What does the application layer mean instead ? which is the other the other component. Application means stopping using models to get answers, but instead starting to design and develop systems that allow us to obtain reliable results. So it means moving from making a call to the large language model to instead starting to orchestrate processes, steps, use of tools, checks to get a reliable output. Andrew says in a beautiful quote: "The hard part of creating AI systems is not using models, because we all have models available and at almost zero cost, but the hard part is creating systems that are reliable in the real world within real processes.

And so I wanted to tell you about two examples of systems we've created for our clients. The first concerns quality translations . We've created a system that allows you to translate brand content into any language. And you're probably wondering, what does this system have that Google Translate doesn't have? Is it a system trained to use the correct terminology for the industry?

It's a system trained to express itself as the brand would express itself. It's a system trained to use a brand-specific glossary in different contexts. It's a system trained not to use terms that are prohibited for the brand, but to replace them with terms that are more appropriate for the brand and, above all, it learns from mistakes that it won't repeat. Sometimes I say it's a bit like having a team of native translators who They work for us within the company, and I don't use the word "team" randomly because the system's infrastructure is more or less like this: it's an autonomous system that retrieves the content to be translated from a private cloud and then activates a multi-agent workflow where different AI agents cooperate to achieve the goal, which is a quality translation for the brand. Then everything is relocated to the private cloud to be processed in other areas.

This is an example of the output. Here you can see the translations of different product sheets for e-commerce in different languages. This system is currently developing around 50,000 translations of 1,000 expired products per day in up to 10 languages. Another example of a system is what I like to call a hybrid editorial team. We developed this for a major publishing group, which is a system that allows you to create, update, and maximize editorial content.

What do you mean by "create"? In the sense that an agent system constantly scans the web and other resources in search of interesting trends for the users. Those trends become an editorial plan. The editorial plan is presented to the publisher, who can approve it partially or fully. The approved editorial plan moves on to another multi-agent workflow that can develop quality content that is saved directly to the publishers' platform.

Finally, publishers can access the platform and finalize the content before publication. What does updating mean? Editorial content needs to be updated over time; it becomes obsolete, or it needs to be optimized over time because it loses performance. So we have an agentic system that is constantly connected to Web Analytics data and constantly extracts the content that needs to be updated and/or optimized. That content is passed to a multi-agent architecture that takes care of updating and optimization, and so what we get is a completely up-to-date system.

We're talking about tens of thousands of pieces of content here because we're talking about publishing platforms of a certain caliber, and then we said maximize. What does that mean? Publishing platforms often have affiliate agreements with the most important marketplaces. So what do they do? They put the products inside editorial content because they then earn money from the clicks that come from users.

But very often these products go out of stock, or the next product model comes out, or they perform poorly within editorial content. So an agentic system monitors all these metrics and is able to intelligently update product banners within editorial content. These examples show us how this union of these two elements— thinking about the first and application layers—allows us, as we were saying before, to reinvent processes with AI at the center, and truly allows us to reach the point where AI performs the most onerous and repetitive tasks, while humans can focus, excuse me, only on quality. And we've also seen that reinventing processes with AI at the center truly changes company flows and truly changes people's roles . I told you earlier, from proof of concept to service to product.

This is an example of the product, it's called Agency, we'll present it next week, and it's a platform dedicated to digital marketing with various tools for digital marketers where each tool— guess what? Where does it come from? It comes from the experimentation that took place within the various teams within the company. Here's an example. You see, here I created an editorial plan for the brand, it's a bit like the agentic flow I showed you before.

I created the editorial plan that shows which contents to produce, which search queries are intercepted, the editorial notes, and so on. Here I can decide to develop one or even all the contents, and the agentic system starts creating the contents. What I get is perfectly optimized content for online performance. So from proof of concept to service to product. Okay, I'm coming to the end, but I'd like to open and close one last parenthesis to make you understand how important the moment of growth we're experiencing is.

About two years ago, this image, which was meant to be ironic, spread online. There was a building under construction with a sheet covering the scaffolding bearing a huge phrase from the series "E c GPT finish this building" from the series. Okay, these systems allow us to produce digital content, but then who... Will the building be finished? But today, the developments in what is called physical AI applied to robotics have allowed us to already be here.

This is specifically an Nvidia project, and when we see this, these applications, that image immediately becomes much less ironic. Do you understand what I meant when I said earlier that we are in a popcorn moment? That is, we are in a moment—and no spoilers, but do you understand what I meant when I said that we are in a popcorn moment, that is, of exponential growth? Well, Demy Sasabis, you've seen it by now, from Google Tip, says that we are experiencing something that is 10 times bigger than the industrial revolution, but the most interesting thing is the second part, that is, that it is going 10 times faster than the revolution of the industrial revolution. Well, then I leave you with the concept I would like you to take home today, which is this: Integrating AI doesn't just mean implementing a technology, but it means embracing a new way of conceiving work and innovation.

And so far, we're on the same page, but the second part is the most important, that is, we are all called to Rethinking our role alongside these systems, because technological innovation will not stop, and what's in the pan will explode. It's up to us as beings, as individuals, as companies, as institutions, as technology leaders, to use the pan lid correctly so we can enjoy popcorn comfortably sitting on the couch instead of spending the entire evening cleaning the kitchen. Thank you.