Between Mountains and Models in Marbella

Since I returned to academia after working in the software industry to pursue my PhD, I have learned a lot. Of course, I learned technical things since I started doing research in two largely new fields (to me): Software and Systems Modeling (within the MDE umbrella) and Machine Learning. I gave much stronger emphasis to the first one, but nowadays I can understand the technicalities and backbone of AI systems much better.

On the modeling side, I quickly learned two things: 1) the MDE community wants to use and promote automation. Ensuring correctness is essential, but automating activities is also a must. It’s something a lot of MDE research is trying to achieve, and 2) the MDE community (at least) really enjoys organizing its activities (meetings, conferences, workshops, etc.) around the sea. Beach locations are frequently the choice for these events.

So, last week I attended the 2026 edition of MODELSWARD (officially the 14th International Conference on Model-Based Software and Systems Engineering), and I confirmed both lessons I’ve been learning since the beginning of my PhD.

The conference was in Marbella, Spain, and just arriving there, the first thing I noticed was the amazing sea coast, of course, but also the beautiful mountains of the city (it’s worth noting that we’re at the end of winter). I have no particular relation with mountains, at least the physical ones. However, I quite frequently use a metaphor to quickly explain my feelings about research at the crossroads of AI and MDE.

To me, these two fields are a single mountain with two sides to be walked up to reach its top. On one side I have MDE, with its strong emphasis on automation from the ground truth (the models) to the actual product (or process, or pipeline, you get me). In contrast, on the other side, we have this computer-powered math and statistics that discovers relations where we were not expecting to find them, ending with some product (usually a prediction or classification, but generative outputs like code, documents, and designs are now a well-established category too). In the end, we want the same thing: better products and processes in the most automated way possible. We can reach that by going up either side of the mountain, or by trying to discover our own path that moves between both sides from time to time, which is my case.

Coming back from Marbella and reflecting about it, I can see that I’m not alone in this perception. A good portion of the papers presented at MODELSWARD were at the same crossroads of modeling and AI. We are trying to solve many MDE problems using AI and, at the same time, leverage the strengths of models to better ground and explain our statistical predictions.

From the specialists panel that opened the conference, to the keynotes, and across the sessions, automation through AI (and, of course, surfing the hype of agents) was there, even if sometimes as a side topic.

In a landscape where machine-speed automation is the baseline expectation and agents are everywhere automating tasks, where do models fit in? As a reliable source of truth to ensure correctness and provide some level of explainability, which we are still missing on the agents’ side.

This was my first presentation at a conference since I finished my PhD, and I believe the future may be as bright as a sunny village on the coast of Spain, but some mountains still need to be scaled. To do so, I think there are many possible paths. What seems increasingly clear to me is that progress will come from connecting both sides of the mountain: combining the push for automation with the need to ensure correct, explainable, well-designed, and well-documented systems.

I’ll keep going in this direction.

Visit the conference website.

Here is the slide deck of my presentation on MODELDSWARD 2026 for those curious about it.

The paper should be released soon with the conference proceedings and I will link it here when it becomes public.