TypeSafe AI unveils Jev, a decision-focused model for software automation

TypeSafe AI launched Jev, its inaugural AI model designed specifically for software decision-making rather than chat. The company, created by ex-OpenAI researcher Diogo Almeida together with Erik Gafni and Sasha Sheng, describes Jev as a “System One Model” that produces structured outputs such as classifications, scores, routing choices and verification results, complete with probability and confidence data. Unlike traditional large language models that generate free-form text, Jev returns predefined, type-safe responses that can be directly consumed by applications.

The model’s name references William Stanley Jevons and the Jevons paradox, underscoring its aim to make AI-driven automation more efficient. TypeSafe is opening early access and seeking developer feedback on use cases ranging from priority tagging to content routing. The firm positions Jev as a complementary layer to larger LLMs, handling the rapid, repetitive decisions that underpin many automated workflows.