In recent years, following the evolution of artificial intelligence has meant following a technology race: every launch brought new comparisons, new benchmark tables and new predictions about which company would lead the next great digital transformation. The whole debate seemed to boil down to one question: which is the best model available today?

Paulo Silveira, co-founder and Chief Vision Officer of Alura, came back from a recent immersion in Silicon Valley – almost ten years after his first visit – feeling that the question is still relevant, but no longer the central one. In such an influential ecosystem, he warns, there is a risk of mistaking consensus for certainty.

From the model race to real-world application

Contrary to much of the public debate, few conversations revolved around the contest between models. Companies, investors and experts were far more interested in understanding how AI is being built into products, processes and day-to-day operations.

In conversations with executives from Snowflake and Elementum, for example, the focus was the transformation of enterprise software: we are entering a phase in which intelligent agents interact directly with large volumes of data, carrying out tasks that today require navigating systems, analysing reports and making a series of operational decisions by hand. For decades, enterprise software was built to organise information; now it is expected to act on it.

Not everyone is heading in the same direction, though. While much of the industry keeps betting on ever-larger models and ever more robust infrastructure, Liquid AI is developing compact systems designed to run locally on phones, vehicles and embedded devices – a reminder that innovation does not move in a single direction, and that the challenge is often to build solutions that are more efficient, accessible and easy to integrate.

Many races, not just one

Seen from afar, AI looks like a single race; up close, many are running at once: large general-purpose models, specialised systems, multimodal applications combining text, image, audio and video, enterprise solutions for very specific problems, and companies betting on making it all lighter and more distributed. The idea of a single dominant technology therefore looks less and less likely.

This also changes how organisations need to think about adopting technology. It is no longer enough to find the best tool and deploy it: integration, governance, privacy, speed of implementation, operating cost and business fit become as important as the sophistication of the technology. A simple solution that is well integrated into processes can create far more value than an extremely advanced technology used in isolation.

In the end, it’s about who uses it best

Conversations about technology, the author notes, often turned into conversations about people. At Stanford, at the heart of one of the world’s most advanced AI ecosystems, much of the discussion focused on education, talent development, organisational culture and leadership. Access to technology tends to become democratised; knowing how to use it strategically remains an advantage that is hard to replicate.

Implementing AI consistently means rethinking processes, building skills, revising workflows, creating oversight mechanisms and preparing people. The next phase of artificial intelligence, Silveira concludes, will not be defined by a single revolutionary model, but by the combination of different technologies and by organisations’ ability to turn them into concrete results. The key question is no longer who has the most advanced AI, but who will be able to use it best.