🔊 How NOT To Build Voice Agents
Details
How NOT To Build Voice Agents
Voice agents in 2026 demo beautifully and fail strangely.
They cut users off mid-sentence.
They treat polite “mhmm”s as interruptions and hang up halfway through a call.
They commit to a phone number after the third digit, then spend the rest of the conversation trying to recover.
The instinct may be to blame the model, but there's more to the story.
Most voice agent failures are not model failures - - They happen in the layer engineers actually control: turn-taking, state management, observability, and the separation of human signal from human noise.
Voice isn't just rendered text to speech input . . it is a different engineering problem with its own failure modes.
This talk walks through four real failure patterns, demonstrates each one with running code, and offers a practical framework for building voice AI agents that survive contact with real users.
Speaker: Rod Rivera, Developer Relations at Rasa.
Rod Rivera leads developer relations at Rasa, the conversational AI platform used by enterprises to build production voice and chat agents.
He has more than 15 years of experience in applied ML, teaches AI Product Engineering at Nebius Academy, and serves as professor of the practice at ITAM in Mexico City.
Rod also writes about the discipline of building reliable, always-on AI agents at sovereignagents.co.
Details:
📍 Location: Improving Plano: 5445 Legacy Dr #100, Plano, TX
🕕 Time: Wednesday, May 27th · 6:00 – 9:00 PM
💻 Livestream Link
Agenda:
6:00 – 6:30 PM
Check-in, networking & pizza
6:30 – 7:30 PM
How NOT To Build Voice Agents
Rod Rivera
7:30 – 8:30 PM
Informal networking
🌐 Stay Connected
If this is your first time, welcome! Be sure to join the group, follow us on LinkedIn, and join our Slack channel to stay in touch between events.