Jev highlights the split between chat models and decision models
The author argues Jev represents a new kind of model that doesn't chat or write code but takes state and a set of allowed decisions and returns typed answers with probabilities, suited to routing, filtering, scoring and moderation.
I think AI is about to split into two very different kinds of model. Models that talk to humans, and models that make decisions for software. Jev is the first thing I’ve seen that made that distinction really click for me. It doesn’t chat and it doesn’t write code. You give it some state and a set of allowed decisions. It returns typed answers and probabilities that your program can act on directly. Sounds less impressive than a giant reasoning model until you think about how much software is basically "given this messy situation, which of these five things should happen next?" Routing, filtering, scoring, moderation, fraud checks, tool selection, escalation, workflow decisions etc etc We’ve been using models built to talk to people for jobs where nobody actually needed them to talk.