Anne Laliron & Arthur Mensch - Italian Tech Week 2025

Hello. So pleased to be with you today here at this great event. So we started our collaboration. It was about 18 months ago and we are here to share our experience. But maybe we can start with the beginning and how we started our collaboration on our side. We were looking for a solution to improve our processes to embed AI in our day-to-day operations. And we found 18 months ago a very inspiring company really helping us and also listening to our pain points and able to collaborate with us to resolve those pain points. Can you tell us more about how it began?

Sure. So for those who maybe don't know what we do at Mistral, we are the only frontier company operating in Europe. It's a company we created two and a half years ago. And really the first thing and the only thing we knew how to do when we started was to train models and that's what we did. That's how we pretty quickly became the leader in releasing open source models. And our starting point at Mistral has always been to create frontier AI and then bring it to everyone's hands. Which means that we are taking a point of view where every enterprise should actually customize to a very large extent the AI technology to make it usable and to make it bring return on investment. And so what that meant is that it was pretty quickly both a science endeavor that is still very core to our business. We have 100 scientists that are working across the globe and we're hiring actually. So please come to see me at the end if you want to join. But the second part which has always been very key to what we do is to work with enterprises. Because in a way we're reaching a point where the horizontal technology kind of looks the same across the different providers. But what really brings the ROI is to go in depth and to look at the use cases that are going to transform the core business of our customers. And so pretty naturally we got in contact and then we started to work on two kinds of problems. One of them really on the operational side equipping people fixing cars, equipping also car assistants. And so we started small, that's how you start as a startup. You start working with the innovation engineers and you start to build trust. And what we're announcing today is something that is actually much bigger and a consequence of the realization that we've been able to achieve in the last 18 months.

Yes, two things. So we are going to be empowering the innovation lab where we're going to be working on again two things: product side, process side. Equipping the sales and post-sales, which is typically the area where we start with enterprises, and we're going to equip the transformation factory as well. So that's the part of Stellantis which is actually delivering the use cases in production, so that they start bringing AI. So for us it's always, I guess, the way we always proceed with enterprises is that there's a discovery part, understanding the use cases that will be effective that go beyond the off-the-shelf solutions, the chatbots that you can buy where typically you need to automate end-to-end processes or you need to rethink about how to organize the teams around AI agents. And then of course the problem is always to take prototypes which are easy to do in AI and move them into production. And that's where you run into edge cases. That's where you have a lot of knowledge transfer to do because it's a different thing to build applications with AI than to build applications with code. And so the two parts, discovering and then going in depth, going all the way to the return on investment is the thing that is distinguishing ourselves.

What we try to do from a macro perspective is to really find champions in multiple industries. Because when we say that AI should be in everyone's hands, it means that the adaptations that needs to be made across multiple sectors are going to be quite different. There is some strong expertise, some decades of IP that somehow needs to make it to the AI systems so that the true value can be unleashed. And for us, Stellantis is our first partner in the automotive industry. And as you said, the automotive industry has all of the characteristics that makes generative AI useful for two reasons. One of them is that it's very operation heavy. As you know there's some manufacturing, there's a very complex bill of materials, very large supplier networks. Also complex sales process and then post-sales process because you need to maintain, you need to make sure that you have the strong quality control. And so all of these things are actually, there's a lot of transactions happening there. You need to talk to your supplier, make sure that they're on time. You have thousands of people that are actually working on those operations and what AI actually allows to do if you think about how to automate certain of the processes entirely is that it's going to save time from those operators so that they spend more time on other things that have more value. So in a way, you mentioned it to me a couple of minutes ago, the automotive adopted the lean approach with Toyota a couple of decades ago where they focused on making everything more efficient physically. And that would typically entail looking at what people were doing manually and really in the physical world. What AI now brings to the automotive industry is this ability to go in the digital world and look at all of the inefficiency in the processes that you could actually automate and do it in a way that is robust because you handle all the edge cases and in a way that can then scale and bring you very meaningful improvement.

Yes. So again there's always when we work with Stellantis and with other enterprises there is always a lot of use cases on the production side and operation side but also a lot of use cases that are a bit more complex to deploy and that typically requires some strong partnership in between our teams that we embed within the teams of our partners and that requires, in general that means bringing AI into the products themselves. And so in the case of cars really you should think about cars as robots and machines that are interfacing with humans. So you have a wheel, you have the pedals, less and less pedals I guess and less and less wheels. But the question there is how are you going to be interacting with the machine in the future, in the next five to 10 years let's say and the answer is pretty clear. We are now able to speak to machines and command them through voice in a way where they understand deeply our intention and can actually interact with complex systems. And so what that means is that we're going to move toward more and more automation and more and more interactive experience driven by voice because when you're driving you can't really do anything else and you can't type, you need to speak. And so the customer experience within the car is going to be very voice-driven and will have a level of personalization that it didn't have before. So on the consumer experience this I think is the most important side of things. And of course a car needs to be maintained. You will continue to be in contact with the garages, with the people that are fixing. And so there the effectiveness of fixing cars and maintaining things is going to be much higher thanks to AI technology because that allows to detect anomalies and then fix them across time.

I would say generative AI overall is about simplifying things. In a way when you used to work on computers and have complex interfaces when you're designing cars for instance, PLM, a lot of different software, kind of clunky because you need to have a lot of options. You used to need to have a lot of options but the power of generative AI is that it understands your intention and it can actually generate the right interface for you to interact with it. So when you're on a computer it means that you start talking to the machine either typing or talking to it and then it shapes an interface for you that is interactive and that's going to help you learn or help you buy things or help you plan ahead. When you're in a car, at the end because the machine can understand the input and human input you can reduce the number of controls to voice pretty closely. And so that means that you need to have less buttons and you can have more automation there. So in general I think the AI technology has to be understood as an interface change where everything that we used to do in a complex way will be able to be done in a more simple way because the machine can understand the intention and adapt the interface to the intention.