TypeSafe: typed decisions inside software

TypeSafe is an existing example of treating model output as a software decision. The useful investigation is where a bounded classification improves a workflow, rather than assuming every task needs a conversational answer.