First software 3.5 model?

Potentially an exciting development in the world of AI - Typesafe AI with the oddly named Jev model - seems to be the first serious attempt at what I've previously called Software 3.5.

What I meant by that was that - as we shift to agentic tasks - i.e. having the robots act as AI oracles in large state machines - it stops making sense to use text based models; we'd be better off with models that are aware they are running state changes. Jev - the first of what Typesafe calls System One models - seems to take a step exactly in this direction.

It's very hard to judge if they are succesful without a paper, evals or access to the model - but this seems directionally interesting.

And about the name: Jev supposedly comes from the Jevons Paradox (then why not just Jevons?) - but show me a single person in the universe who speaks English, sees 'Jev' and does not instantly pronounce it like 'jew'.