After an initial voracious, generalist phase, AI enters the era of specialisation

Summary

This opinion piece argues that competition in AI is shifting from the race for general-purpose models towards industry-specific specialisation. It cites the concept of DSLM (domain-specific language model) introduced by Gartner, presented as the only viable approach for critical industries (cybersecurity, banking, insurance, defence), using fine-tuned models hosted on premise to ensure data sovereignty. The author refers to several cases: Anthropic's restriction of access to its Mythos vulnerability detection tool, a non-public OpenAI model that hacked Hugging Face, SpaceX's acquisition of Cursor, and Meta slowing its infrastructure build-out in the face of the cost of AI agents. Safran Aircraft Engineering is cited as an example of a company that has deployed a multi-model architecture, reducing the processing time for regulatory certification files from several weeks to a few minutes. The article argues that companies that remain stuck in a fixed use of AI on closed cloud platforms are depriving themselves of the most advanced agentic capabilities.

Summary based on the full text of the article.