AivexaNewsSearch
AI news for builders and product teamsChecked every hour

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

Collected Oct 11, 2026

Executives working on voice AI said the technology has not yet reached the breakthrough moment that chatbots experienced with ChatGPT, even as investors pour money into the sector. Speaking onstage at the HumanX conference last month, PolyAI CTO Shawn Wen said the industry has cleared one major technical hurdle but faces another. Otter CMO Alex Gay joined the same discussion.

PolyAI is an enterprise voice AI platform, and Otter makes a meeting note-taker. Their remarks came as voice AI attracts billions in investment across model makers, enterprise customer service providers, meeting note-takers and AI-powered dictation tools. New models and tools arrive weekly with claims of human-sounding conversation, but the executives argued that reality lags the marketing.

Wen framed the current state as a two-part problem. Full-duplex models, which can speak while listening to the user, have already been developed. What remains, he said, is making reasoning fast enough that models can fetch answers quickly and keep a conversation feeling natural. That is a latency and retrieval challenge as much as a speech challenge. In practice, a voice agent has to recognize what was said, decide what it means, look up or generate the right answer, and speak it back, all inside the pauses a human caller tolerates. Each step adds delay, and the slowest step sets the pace of the exchange. This is my reading of the trade-off the executives described, not a claim in the source.

Wen also said customer service agents should not sound robotic and should give callers confidence that their problems can be solved. He described the shift as gradual: once the voice quality is good enough and a customer engages for the first two or three turns, confidence builds, and over time the customer may feel they do not need to talk to a human if the agent can resolve the issue.

Gay pointed to a different part of the stack. He said speaker identification, intent capture, and writing that up alongside organizational knowledge are key steps toward enabling automation. Otter is working on digital twins that could represent people in meetings. For that, he said, the output voice needs to convey the same emotive expressions as talking to a human in a meeting. He compared the best meetings to debate and strategic discussion underpinned by a relationship, and said that without that quality an avatar is just a question-and-answer chatbot.

Both executives flagged accuracy problems in the underlying speech layer. Wen said ASR, or automatic speech recognition, models often miss important keywords, which then breaks context capture for the whole pipeline. Gay agreed, saying Otter keeps working on improving transcription and that language is an area where voice models need to improve.

Gay explained why transcription accuracy matters so much for a note-taker. Transcription was never the end point for Otter, he said, but the layer on which productivity gains are built. If the original transcript lacks the needed accuracy, every follow-up action built on it becomes flawed. Once the system starts acting on a wrong transcript, trust in the platform is lost. He called continued improvement of the ASR model critical because the downstream impacts are significant. That chain is easy to see: a missed name in a sales call becomes a wrong action item; a garbled requirement becomes a wrong ticket. My inference is that vendors selling automation on top of transcription are only as reliable as their weakest recognition step, which makes ASR performance a competitive battleground rather than a solved commodity.

Transparency was the other theme. The executives said tools should tell customers when they are being recorded or when they are talking to an AI. Otter said it wants to instill trust in meeting participants, and even when its bot is not present it wants to try methods such as notifying everyone in the chat that the meeting is being recorded. Wen said it is important to establish that people are talking to an AI on enterprise calls.

Why it matters: Teams building voice agents, meeting assistants and customer service bots should treat fast reasoning and ASR accuracy as the gating factors rather than model polish, since the executives tie user trust to both. Product teams in particular will face disclosure expectations, because the people in these calls are meant to know when software is listening or speaking. Developers weighing voice interfaces against text should read the source's message as a caution: full-duplex conversation is available, but natural, reliable interaction still depends on latency and recognition quality that are not yet solved.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Voice AI often misses important points for its context layer, and causes the whole pipeline to break.