Voice AI Agents Decoded | Plivo, Deepgram & Inworld Leaders on What Breaks and What Works
Voice AI agents are everywhere. Production-ready ones aren't.
Building a voice agent that actually works means wiring together speech recognition, language models, text-to-speech, telephony, and orchestration - then tuning for latency, handling edge cases, and hoping it all holds up at scale. The founders and product leaders solving these problems at each layer rarely end up in the same room.
Plivo is hosting an evening in San Francisco with builders from across the voice agent stack for a conversation about what it really takes to ship voice agents that work in production.
Join us for a candid, unscripted evening with founders and leaders from across the voice AI stack:
Vyas A - Head of Product, Plivo: The Voice AI platform to build and deploy human-like voice agents.
Kylan Gibbs - CEO, Inworld: The platform powering lifelike text-to-speech for interactive AI experiences.
Anoop Dawar, CSO, Deepgram: The fastest, most accurate speech-to-text API for real-time voice applications.
What We'll Get Into
The Latency Chain — Every millisecond compounds across STT, LLM, and TTS. Where the bottlenecks actually live today, and how each layer is attacking them.
Build vs. Buy vs. Stitch — End-to-end platform, best-of-breed pipeline frameworks, or build from scratch. The real trade-offs behind architectural decisions at scale.
What Makes an Agent an Agent — Beyond basic Q&A: tool calling, multi-turn memory, mid-call routing, and the features that separate a voice bot from a voice agent.
The Production Gap — Demo ≠ deployment. What actually breaks in the wild — noise handling, concurrency, hallucinations, failure recovery — and how teams are solving it.
Where This Is Headed — Full-duplex speech, multimodal agents. The bets these leaders are making in the next 12 months.
What to Expect
5:30 PM: Networking
6:00 PM: Panel discussion + Audience Q&A
7:00 PM: Food, drinks & networking