Washington just proved frontier AI is a national-security asset — and your enterprise stack is the collateral.

paid for an agent company with a model that beats Claude on its own turf

Fin runs on its own model, Apex, which the company says it post-trained specifically for support and which it claims outperforms frontier models from OpenAI and Anthropic on resolution. Salesforce did not buy a wrapper. (Salesforce investor release)

Four of today's five stories are arguments about where AI value actually sits, and they don't agree. Washington proved the model layer can vanish on 90 minutes' notice; OpenRouter showed a judged panel of models beats any one of them; Salesforce paid $3.6B for a post-trained vertical model the same day a Nature Medicine study found general models out-answer the specialized clinical tools hospitals already license. The through-line is not "vertical wins" or "frontier wins" — it is that the single-model bet is the fragile one, whether the fragility is regulatory, competitive, or just a better ensemble down the hall. The durable position is the layer that survives a model swap: orchestration logic, workflow integration, and the regulatory relationship. The bet quietly breaking is the one 2025 strategy decks treated as bedrock — a business pinned to one frontier API and the vendor behind it.

— Agentic desk