Cerence’s non-linear upside is not “more cars,” it’s “more dollars per shipped car + per active driver” as assistants shift from Q&A to doing safe actions (settings, navigation, service, commerce). With software commoditization rising, the defense is to price what generic models can’t: embedded workflow integration, auditability, and liability-reducing verification for high-risk actions. If Cerence converts
xUI programs into production launches and standardizes outcome/usage packaging, the mix should look more recurring and less IP-lumpy, supporting a partial
multiple recovery from today’s distressed level. Agent disintermediation risk is contained only if Cerence becomes the required action gateway (policy, authentication, logging) rather than just the UI.