The case against enshittification: why specialized, sovereign ai beats generic pilots every time

In a post this week, German AI lab Aleph Alpha takes a hard look at the state of enterprise artificial intelligence. It borrows the term "enshittification", coined by writer Cory Doctorow and surfaced at the Gartner IT Symposium/XPO in Barcelona by analyst Gabriela Vogel, to describe a downward spiral where platforms sacrifice quality and trust for growth and engagement.

"For serious work, a self-confident failure costs more credibility than ten successes can bring," the company writes, quoting practitioner Tanmai Gopal. The point lands hard: when systems promise reliability and then stumble, the damage to organizational confidence is swift and lasting.

Founded in 2019, Aleph Alpha positions itself as a European champion of sovereign AI, a stance the company insists is not a marketing slogan but a technical strategy. The post contrasts generic pilot projects, which the company argues "promise too much and deliver too little", with a narrower, more disciplined approach: specialized, domain-specific AI agents built through co-creation with customers.

Sovereignty as trust infrastructure

The company frames sovereignty along three dimensions: data sovereignty, keeping intellectual property under the customer’s control; technological sovereignty, avoiding black-box intelligence; and operational sovereignty, eliminating platform lock-in. The argument is that this triad turns data from a liability into a strategic asset.

To illustrate the point, Aleph Alpha recounts a live session at the Barcelona conference, Why AI Pilots Fail: Generalists Don’t Succeed, Specialists Do, presented by VP of Community Sven Körner. The session featured a customer whose engineering team faced massive volumes of reports, fragmented systems, and highly technical documentation. Instead of a generic large language model, the team deployed domain-specific AI agents that combine a knowledge graph with neural networks, enabling autonomous analytics and administrative tasks, merging and contextualizing problem reports across systems.

"Our customers entrust our sovereign AI platform PhariaAI with their real operational data and intellectual property. They commit to sharing their expertise and processes so that we can build a solution that fits them perfectly."

The result, according to the company, is a production-grade system that operates at the level of human experts in its specific domain. The customer was reportedly surprised at how effectively AI could simplify complexity, discovering use cases far beyond the original few they had envisioned.

The cost of credibility

Aleph Alpha is making a broader industry argument: one-size-fits-all models fail not because the technology is unready, but because they ignore what makes each organization unique. When they fail, the company warns, they don’t just stall a project, they break trust. The blog post concludes with a sharp statement on vendor relationships: "When organizations hand over their knowledge to black-box systems, they trade dignity for dependence. For us, that's the opposite of sovereignty. It's complete capitulation, and frankly, an expensive mistake."

The company’s stance echoes a growing sentiment among European AI builders: that enterprise adoption will not scale through generic foundation models alone, but through systems deliberately constrained to specific domains, built on transparent, auditable, and controllable infrastructure. Whether that approach gains enough traction to shift industry practice remains an open question.