These Execs Think Voice AI Hasn't Reached Its ChatGPT Moment Yet

My take: Two executives building voice AI products (one at PolyAI, which builds customer service agents, and another at Otter, which builds meeting transcription) agree on something refreshingly honest: voice AI still hasn't had its 'ChatGPT moment,' the kind of leap that gets everyone using it overnight. Models that can talk and listen at the same time (called 'full duplex') already exist, but they still need to respond fast enough and understand every word accurately enough for a conversation to feel natural.

This strikes me as a more honest read than most of what gets said about voice AI, which is why it's worth sharing. If your business is considering a voice agent for customer service, transcription is the foundation of everything: if the system misses key words, every step after that (the response, the action, the report) inherits that error. The executives themselves insist on something I agree with: if a customer is talking to an AI, you need to tell them clearly, not disguise it as a person.

The real opportunity is using these tools where they already work well today: transcribing meetings accurately, capturing who said what, or giving service calls a first filter before handing them to a person. Waiting for the technology to be perfect before trying it wastes time; launching it without testing its accuracy first risks your customers' trust.

If your business already uses or is considering an AI voice assistant, are you testing it first on low risk tasks, or did you launch it straight into talking with your customers?