These execs think voice AI hasn’t reached its ChatGPT moment yet

Investors have poured billions into voice AI, from model makers to call-center platforms to meeting note-takers and dictation apps. New models arrive weekly promising to sound human. Yet two executives building in the space, PolyAI's CTO Shawn Wen and Otter's CMO Alex Gay, say voice has not had its breakout moment.

Their diagnosis is useful because it is not about voices sounding nicer. Wen says full-duplex models, the kind that can speak while still listening, are already here. The bottleneck is fast reasoning, so answers arrive without the pause that breaks a conversation, and speech recognition that does not drop the keywords everything else depends on.

We think this is the right framing, and it is a healthy one coming from inside the industry. The demo era of voice AI was about timbre and latency. The useful era will be about whether the thing actually heard you correctly and did the right job.

Gay makes the sharpest point. For Otter, transcription was never the product; it was the layer everything else sits on. If the transcript is wrong, the summary is wrong, and once an agent starts taking actions off that summary, the errors compound. That is the real risk in voice automation. A mistyped word in a chat window is visible. A misheard word in a phone call or meeting quietly becomes a wrong refund, a missed task, or a misattributed decision, and nobody notices until it matters.

Wen's picture of customer service is a trust ladder: if callers will stay engaged for the first two or three turns, they start believing the agent can solve the problem, and eventually stop asking for a human. We see the upside. Long hold times are miserable, and an agent that genuinely resolves issues is a better experience for many people. But we would watch who that confidence serves. If companies use good-enough voices to make reaching a person harder rather than unnecessary, the trust ladder becomes a trap.

The digital-twin idea from Otter, avatars that might stand in for people in meetings, is where I'm most cautious. Gay is right that the best meetings run on debate and relationships, and that an avatar without that is just a Q&A bot. The flip side is that an avatar convincing enough to carry a relationship raises harder questions about consent, representation, and who is accountable for what the twin says.

That is why the transparency piece cannot be an afterthought. Wen says enterprise callers should know they are talking to an AI, and Otter describes notifying meeting chats even when its bot is not visibly present. Good. If voice AI scales the way its funders hope, disclosure needs to be a default the whole industry holds to, not a feature a few vendors choose.

Wen and Gay spoke with TechCrunch's Ivan Mehta on stage at the HumanX conference last month. PolyAI builds enterprise voice agents for customer service; Otter is a meeting note-taker that is working on digital twins to represent people in meetings.