August 2026 Summaries

Nathalie Criou, Vapi’s new VP of Product, brings experience spanning engineering, marketing, sales, product management, entrepreneurship, and leadership roles at companies including Google, VMware, Amazon, Twilio, and Docker, as well as her own startup, RidePal. She joined Vapi because she sees AI as an opportunity to restore voice as an accessible, effective business channel after poor customer experiences reduced its use. Criou argues that voice AI already delivers repeatable returns and could dramatically increase customer conversations as companies remove cost-driven limits on phone support, citing customer Kavak’s expansion of voice AI across its business. Her product strategy emphasizes serving both developers and increasingly capable nontechnical users while remaining focused on customer outcomes such as revenue growth and cost reduction rather than treating AI as a goal itself. She advocates organizing product teams around opportunities and problems rather than existing features, believing this encourages teams to pursue better solutions, and she identifies sailing and cats as important interests outside work.

Conversational AI design for voice calls involves addressing real-time systems challenges to create natural and seamless interactions, distinct from the design for text-based chat. Unlike text, where users are more forgiving of delays, voice interactions require immediate responses, precise turn-taking, and handling of interruptions to avoid frustration. Key factors that determine the quality of a voice conversation include latency, turn-taking, interruption management, prosody, and a resilient, model-agnostic pipeline that can adapt to provider outages. Vapi emphasizes a flexible architecture allowing for custom tuning of these dimensions, ensuring the voice agent sounds human and can navigate disruptions during a call. Testing voice agents through simulated calls before deployment is crucial, as it helps identify and rectify potential issues in timing and delivery, ensuring high satisfaction and effectiveness in real-world conditions. Vapi facilitates this process with a low-code platform, enabling users to configure and refine voice agents without extensive coding, thereby improving metrics like Net Promoter Score (NPS) and conversion rates.

Recent developments in the Humanness Index™, a benchmark for evaluating how human-like voice models sound, highlight the rapid advancements in synthetic voice technology. Two new models, Speechify's Simba 3.2 and Fish Audio's S2.1-Pro, have recently topped the leaderboard, with Simba 3.2 achieving a score of 99, nearly indistinguishable from a human voice. These advancements underscore the importance of using flexible voice platforms that allow for swift updates to leverage the most advanced models available. This fast-paced innovation means that models that were top-ranked just a few months ago have now been surpassed, illustrating the dynamic nature of the field. The Humanness Index™ continues to evolve with thousands of votes from listeners who compare synthetic voices to real human recordings, ensuring that the benchmark remains current with the latest technological progress.