Voice AI executives are openly acknowledging that despite billions in venture funding and weekly model releases, the technology remains fundamentally immature for enterprise deployment. PolyAI's CTO Shawn Wen articulated the core problem: full-duplex models—systems that can speak whilst listening—represent a solved technical problem, but the real bottleneck is inference speed and reasoning quality. When customers experience lag between their question and an AI's response, the illusion of natural conversation collapses, and with it, the willingness to treat the agent as genuinely capable. This mirrors the broader pattern across voice AI platforms: Otter's leadership confirmed that transcription accuracy remains the critical dependency, with downstream automation becoming unreliable the moment ASR (Automatic Speech Recognition) models miss contextual keywords. The implication is stark—teams currently piloting voice agents in contact centres should audit whether their underlying transcription and reasoning layers are genuinely production-ready, or whether they're building automation on a foundation of compounding errors.
The transparency and trust gap presents a secondary but equally consequential challenge. Both PolyAI and Otter emphasised that customers must know they're interacting with AI, yet this disclosure requirement sits uncomfortably with the industry's push toward "human-like" interactions. For CX teams already running voice-enabled platforms—whether through Salesforce's Agentforce or specialist providers—this creates a strategic tension: the more convincing the voice agent, the greater the regulatory and ethical obligation to disclose its nature, and the higher the reputational risk if that disclosure feels deceptive. Otter's observation that trust evaporates the moment downstream actions prove flawed is particularly relevant for support teams relying on voice summaries to populate knowledge bases or trigger automations. The real ChatGPT moment for voice AI will arrive only when these systems achieve both technical reliability and transparent trustworthiness simultaneously—a combination that remains elusive despite the hype cycle.
These execs think voice AI hasn’t reached its ChatGPT moment yet TechCrunch