PolyAI CTO Shawn Wen: voice AI lacks its "ChatGPT moment"

Voice AI has drawn billions of dollars in investment, with startups spanning model makers, enterprise customer service, meeting note-takers and AI-powered dictation. Releases claiming human-like conversation arrive almost weekly, which makes the on-stage assessment from PolyAI's CTO notable: the technology has crossed one threshold — full-duplex models that can speak while listening — without yet producing the breakthrough moment that made ChatGPT a household reference point.

Wen locates the remaining gap in speed rather than voice quality. Once the voice is good enough and the customer is willing to engage for the first two or three turns, he says, confidence builds and the caller may conclude they don't need a human if the agent can solve the problem. That sequence turns reasoning latency into a business question for customer-service deployments, not just an engineering one.

He also argued that service agents should not sound robotic — a point that sits alongside the transparency expectations discussed by Otter, whose CMO Alex Gay described accuracy in transcription as the layer feeding downstream automation, and noted both companies' emphasis on telling people when they are being recorded or talking to an AI.