Volviendo (otra vez) al tema de la soberanía digital

The topic of digital sovereignty has resurfaced in recent discussions. It concerns the idea that companies and nations should be able to utilize their own artificial intelligence (AI) trained by themselves, rather than relying solely on the major labs that currently dominate the market. Numerous models of this type already exist, ready for use.

However, the cost remains a significant barrier. The necessary servers seldom match the scale of massive computing centers where AI is trained for production, and even then, they remain expensive. Consequently, only large companies or nations with substantial resources can undertake this challenge.

Interestingly, a middle ground exists: a modestly budgeted company—perhaps spending around $4,000—can operate a machine with its own AI. While this setup has certain limitations, it can still deliver a performance that is more than satisfactory. It seems I am once again revisiting this subject. The reason is that Anthropic's CEO recently announced that their development systems can only use their own company-specific markdown files ( *.md ), namely one called claude.md.

On the surface, this seems trivial, as one might think simply naming the file "claude.md" would suffice. Yet, it is not that straightforward.

For those unfamiliar, a "skill" in this context refers to a text file containing instructions for an AI to know how to operate. While purists may disagree with this simplification, the essence is accurate. The issue arises because other AI systems utilize a file with a different name, AGENTS.md, whereas Claude specifically requires its own.

This means that, depending on the AI agent desired, separate files must be maintained, essentially performing the same function. This situation echoes a historical precedent: when a devilish company, associated with a windowed operating system, created its own browser with a similar concept, ignoring W3C standards and introducing a patchwork of code.

The outcome was filled with patches, duplicated instructions, and web pages that appeared fine on one browser but not, or not at all, on others. This is akin to what Anthropic is proposing now with their proprietary markdown files: we will need different files, each performing the same task, depending on the AI agent being used.

Indeed, some major companies have expressed dissatisfaction, as they possess hundreds or thousands of these files and now must duplicate them to ensure compatibility with Anthropic's agents.

Therefore, I emphasize once more the importance of avoiding such commercial arbitrariness. Digital sovereignty appears to be the best way to prevent becoming trapped by these whimsical technological preferences imposed by a few market players.