Voice AI executives identify reasoning speed and transcription as reliability gaps
PolyAI and Otter executives say natural-sounding speech is not enough when recognition errors can corrupt downstream actions.
Fluent speech does not guarantee a useful agent
PolyAI CTO Shawn Wen says voice AI has not yet reached its equivalent of a ChatGPT moment, despite progress on full-duplex systems that listen and speak simultaneously. In an account published by TechCrunch on October 11, he identifies faster reasoning and answer retrieval as the next challenge for making conversations feel natural. The comments came from a HumanX conference discussion the previous month, not a newly released model evaluation.
Otter chief marketing officer Alex Gay highlights speaker identification, intent capture and the connection to organizational knowledge. He argues that accurate transcription is the foundation for useful automation: if the initial record is wrong, follow-up actions can also be wrong and undermine trust. Wen similarly says missed keywords can break the context needed to handle a caller’s request.
Disclosure is part of the discussion
Both executives also address transparency. Wen says enterprise callers should know they are talking to AI. Gay discusses notifying meeting participants that recording is taking place, including situations without a visible meeting bot. These are the executives’ positions and descriptions of work, not proof that every deployment already provides those notifications. The report does not supply independent accuracy measurements or establish that either vendor has solved these problems.