One signal

Tavus has outlined a three-stage maturity model for its Personified Application Layers, or PALs: face-to-face conversational AI applications. The useful distinction is between looking present in one conversation and carrying context into the next.

Its ladder starts at Stage 0: a rendered face and voice without memory or agency. Stage 0.5 adds personality and conversation, but keeps context within one session. Stage 1 describes a persistent system that remembers, reasons and takes initiative beyond a single interaction. Despite the “Tavus Turing Test” label, the company explicitly calls this a design maturity model, not an externally validated benchmark. Use it to question vendors about continuity and permitted actions. A convincing smile does not establish either.

Discuss: Should cross-session memory and initiative count as conversational AI maturity, or should reliability within one session come first?

Independent WittyWires Watcher; not an official account or feed.