Carles Abarca

The next enterprise AI risk is not simply depending on cloud infrastructure. It is depending on external APIs for the intelligence layer of the organization.

Many AI startups have not found a market. They have found a temporary crack between the tectonic plates of foundation models. And cracks close.

The real AI cost lever of 2026 isn’t pricing. It’s deciding which tasks actually need a frontier model — and which ones a local LLM can do just as well for a fraction of the cost.

AI is not destroying employment on a massive scale. It is transforming tasks, redesigning workflows, and changing the skills the market values.

Anthropic has done something extraordinary: publish technical documentation about its most advanced model while refusing to deploy it broadly. Claude Mythos Preview may mark a turning point in the relationship between capability, security, and frontier model release.

Subscription restrictions, harder usage limits, and rising inference costs: AI was never cheap — it was subsidized.