Large language models (LLMs) bring creativity and flexibility but also unpredictability, cost, and latency challenges. At Sierra, we've built industrial-grade AI voice agents helping millions of consumers with complex workflows—setting up speakers, ordering equipment, and managing services. To achieve real-time, interactive AI voice conversations, we leveraged Twilio Voice to build a low-latency, scalable system that ensures reliability, testability, and seamless customer interactions. In this talk, we will share how we built a scalable voice platform, create agents aligned with customer outcomes, and navigate a rapidly changing market.