Why New AI Models Won't Fix Your Voice Agent | Carter Huffman, Modulate

Spoken words carry only 3–20% of the meaning in a voice conversation - the rest is tone, pace, and emotion, and none of it reaches a transcript-based LLM. Modulate CTO and co-founder Carter Huffman explains why upgrading your voice AI model doesn't fix this: it's an architecture problem, not a model quality problem. Huffman also cites a 2026 finding that over 59% of consumers report frustration with AI voice agents — even as the underlying models keep improving. Watch the full talk, on Ai4's channel: https://bit.ly/3SWlwua Modulate builds audio-native voice AI models for real-time conversation understanding, including Velma, an ensemble listening model for voice agent observability. Get 1000 free API credits to test Modulate's AI models here: https://bit.ly/4gMSTZt