Codestrap uses Jev for state machine programming and guarded transitions
Codestrap applies Jev to state machine programming, control flow and next best-state prediction, and plans to benchmark it against its current Gemini 3.1 Flash-Lite agents.
At @Codestrap_ai we have been experimenting with Jev for programming state machines. At first glance, you might think this is an anti-pattern for a model that cannot generate text. But because our agents work symbolically within a bounded context, valid states are known ahead of time. We use an iterative process you can test out in our dashboard (link in comments). We also use Jev for control flow inside state machine executions. Our next benchmark will compare full Jev based agent executions against our current Gemini 3.1 Flash-Lite agents. Jev will be used for guarded transitions and next best-state prediction, in addition to state machine programming. We believe the results will be pretty amazing for Jev. It’s awesome how well aligned this class of models is with our investments in neurosymbolic programming. @typesafeai .