The Jev Revolution: What Engineers Are Already Building with System One Models — Latent Space

Why it matters

At the AI Engineer Code Summit, sessions centered on system one models — non-generative systems that return structured choices, scores, or probabilities, and that typically run faster and at lower cost than reasoning LLMs. Diogo Almeida, co-founder and CEO of TypeSafe AI and lead creator of Jev, delivered a keynote arguing that Jev represents reliable intelligence suited to automating everything, and that such models must integrate with code because code is the language of automation. Latent Space founder swyx endorsed embedding system one model calls throughout application code to build smarter software. Practitioners from Glean, UiPath, and Workday showed real deployments: zero-shot classification that avoids repeated fine-tuning, a tournament-style method for ranking document chunks in under a second, and interfaces that predict user actions while typing. The story matters because it marks a shift in AI engineering discourse away from generative models alone, toward cheap, fast, reliable components woven directly into ordinary software, which Almeida compared to the experimentation of the early internet.