By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
YC Root AccessWhat It Actually Takes to Deploy a Voice Agent to a Fortune 500
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
At a glance
WHAT IT’S REALLY ABOUT
Scaling enterprise voice agents with simulation, evaluation, and observability infrastructure
- Coval provides simulation and observability so companies can test voice agents at scale before exposing real customers and monitor behavior across tens of millions of production calls.
- Enterprises are adopting voice agents quickly because existing call-center infrastructure (IVRs, SOPs, call flows) makes the jump to autonomous voice less radical than other agent categories.
- Voice agents introduce new failure modes—audio oddities, hallucinations, and compliance/security risks—making rigorous QA and continuous evaluation essential rather than optional.
- The next performance unlock is better controllability for real-time/speech-to-speech systems and improved coordination between specialized components (STT, reasoning, TTS) that share context effectively.
- Coval’s company journey hinged on strong customer pull from voice teams, an intentional early focus on enterprise scale, and founder-market fit from Hopkins’ Waymo simulation/evaluation tooling background.
IDEAS WORTH REMEMBERING
5 ideasEnterprise voice is the first truly scaled autonomous-agent deployment category.
Hopkins argues voice is already “productionized” autonomy, because companies can slot agents into existing phone workflows and immediately run them at high volume.
Testing voice agents with real calls doesn’t scale—simulation is the missing infrastructure.
Manually calling an agent repeatedly is slow, expensive, and statistically weak; simulation enables broader coverage of edge cases before risking customer-facing failures.
Voice QA must expand beyond classic contact-center metrics.
Enterprises need to evaluate not just outcomes, but also workflow/tool-call correctness and audio quality factors like latency, interruptions, and background noise handling.
Word error rate is often overvalued relative to intent and task success.
A conversation can tolerate imperfect transcription if the system correctly infers intent and completes the workflow, similar to humans understanding garbled audio on calls.
Agents fail in more “egregious” and brand-risky ways than humans.
Examples include vocal hallucinations (screaming/whispering/voice shifts) and confidently wrong statements—issues that require new guardrails and monitoring.
WORDS WORTH SAVING
5 quotesCoval is a simulation and observability platform for voice agents, so we help you to scale your voice agents over millions of conversations so that you don't have to test your voice agents with real customers in production, and then also when you deploy your agents to production, you know what's happening out there in the wild.
— Brooke Hopkins
I think in two years or one year from now, it's going to be unacceptable to call an airline and be on hold for 20 minutes.
— Brooke Hopkins
Famously, voice agents will accidentally scream, or they'll start to whisper, or they'll change voices halfway through.
— Brooke Hopkins
It's been really cool to watch Coval grow because now I feel what product market fit feels like, where people are chasing us down to book meetings to push things through procurement. They're putting us on their back and just carrying us through the procurement process.
— Brooke Hopkins
So it's pretty easy to make something work once, but then to make it work over time is the challenging part.
— Brooke Hopkins
SPEAKERS
Harj Taggar
hostPartner at Y Combinator and host of YC Root Access.
Brooke Hopkins
guestFounder and CEO of Coval, focused on simulation/evaluation for production voice agents.
High quality AI-generated summary created from speaker-labeled transcript.
