Decisions, not paragraphs
Large language models are excellent at expression. Software often needs something narrower: route this ticket, block this action, score this lead or choose the next model.
JEV turns those judgments into fixed output types, so code can act directly or ask for review when confidence is low.
System One Models
TypeSafe AI describes JEV as its first System One Model: fast, machine-native judgment designed to live inside software.
LLM vs JEV
NeedLLMJEV
Output shapeParagraphs you have to parse againchoice / score / noul
ThresholdsHard to turn “I'm unsure” into a ruleCalibrated probability you can branch on
Fits in codeNeeds an extra interpretation stepReturns values a workflow can execute
When not to use JEV
Skip JEV when you need explanation, long-form writing, or a multi-turn conversation. Use an LLM there. JEV is for explicit judgments that should trigger the next software action.