While we recently explored how traditional benchmarks are losing their grip on the true capabilities of Large Language Models, the real-world test isn’t happening in a lab—it’s happening in our pockets. We’re moving from measuring raw parameters to evaluating how these models actually converse and integrate into our daily workflows.
At Ambiente Ingegneria, we’ve been tracking this shift toward agentic systems for a while. We’ve previously discussed the “death of the app” and the rise of deep-access AI, but seeing these concepts move from the whiteboard to the halls of the Mobile World Congress is a game-changer. The conversation has shifted: it’s no longer about the handset, but about the invisible assistant living inside it.
We are seeing a new generation of devices, particularly from the Chinese market with models like Qwen3-Max-Thinking rivaling Gemini 3 Pro, that bypass the surface layer to access the deepest parts of an operating system. As engineers, this excites us, but it also raises a red flag regarding digital sovereignty. When we build web applications using Python, Django, and Flask, our focus is always on the data architecture. We believe that for AI to be truly useful, the underlying PostgreSQL or MySQL databases must be engineered to protect user privacy, not just feed a hungry model.
Interestingly, making an AI “sound human” still requires a very human touch. Reports of people earning $600 a week just to talk to AI highlight a crucial truth: nuanced human feedback is the only way to refine RAG (Retrieval-Augmented Generation) and voice interfaces. This human-in-the-loop approach is exactly what we implement when integrating LLM assistants into Odoo ERP or custom mobile apps. It’s about ensuring the AI isn’t just statistically accurate, but contextually wise.
In engineering, we rely on the metric system because it provides a universal, objective standard. We need the digital equivalent for AI—standardized metrics that cut through the marketing hype of “thinking models” to ensure they are safe, reliable, and free from the noise of fake news. Whether we are developing a custom Odoo module or a React-based front-end, our goal remains the same: using clear standards and quality data to build tools that people can actually trust.
Source: https://www.abc.es/opinion/sevilla/gustavo-fuentes-despues-movil-20260312203834-nts.html
