August 2026 Summaries

Vapi has launched a rebuilt version of Campaigns for batch outbound calling, enabling users to call contact lists of up to 10,000 people with personalized assistants or squads and configurable phone numbers, calling windows, and conversion checks. The new system replaces an earlier batch-based design with a durable workflow engine that preserves campaign state through restarts or failures, manages concurrency through a sliding window of active calls, and records immutable events for each campaign and contact to provide accurate progress tracking and audit trails. Contact data can be inserted as prompt variables, while pre-dial webhooks can consult external systems to skip contacts who have already converted or avoid calls when webhook errors occur. Campaigns also provide live status visibility and allow users to stop new dials while active calls finish. Vapi positions the upgrade for operational use cases including healthcare appointment reminders, payment collections, sales lead qualification, and staffing outreach, with integrations supporting real-time CRM or system updates.

A healthcare scheduling software company’s attempt to build an in-house voice agent illustrates how a quick demo can expand into months of unplanned work on real-time audio, telephony, model failover, state management, observability, interruption handling, and reliability at scale. The passage argues that voice AI is fundamentally more complex than adding audio to a chat workflow because conversational latency, endpoint detection, carrier behavior, and concurrent-call performance require specialized, ongoing engineering. It suggests that platforms can reduce delivery time by providing infrastructure, integrations, provider flexibility, and operational expertise, allowing product teams to focus on conversation design and core business workflows. Building internally may still be appropriate when a company has unique pipeline requirements, substantial existing voice infrastructure, strict data-processing constraints, or voice technology as a central competitive advantage, but for most SaaS companies treating voice as a product feature, the recommended approach is to use a platform, validate customer demand quickly, and avoid diverting engineering resources from the primary product.

AI voice agents for customer experience should be evaluated by their ability to resolve customer problems rather than merely contain or deflect calls, particularly as consumers increasingly prefer phone support for complex issues. The proposed seven-part evaluation framework assesses resolution, real-time voice quality, reliable action-taking in business systems, context-rich escalation to human agents, model flexibility and failover, lifecycle testing and observability, and the degree of platform control available to customers. Because voice interactions have far lower tolerance for latency, interruptions, background noise, and awkward turn-taking than chat, production systems require tightly coordinated speech recognition, language models, and speech synthesis. The market includes packaged horizontal applications, vertical point solutions, contact-center-native add-ons, and open platforms, with the choice depending on whether organizations prioritize rapid deployment or ownership of their workflows, data, model providers, and ongoing improvements. Vapi is presented as a model-agnostic open platform that supports tool integrations, configurable call behavior, logging, testing, evaluations, and warm transfers, while the broader recommendation is to start with a narrow, measurable use case, test against real calls, review failures regularly, and iteratively expand scope as resolution improves.

Vapi Simulations is a native AI-powered testing feature for voice agents that uses configurable AI callers with distinct personalities and realistic scenarios to hold conversations with agents before deployment. Each test produces pass-or-fail results, transcripts, recordings, and structured evaluations, helping teams identify conversational failures and regressions that manual testing or scripted evaluations may miss. Users can create personalities, scenarios, individual simulations, and reusable test suites, while tool mocks enable testing of API errors or timeouts without affecting live systems. Chat mode supports rapid, lower-cost text testing, while voice mode validates the complete audio experience, and API integration allows quality gates to block deployments when test performance declines. The feature is intended to help teams validate launches, prevent regressions, verify compliance guardrails, and test difficult failure conditions such as unavailable appointments or calendar-service failures.

Enterprise buyers evaluating conversational AI for production voice agents should prioritize reliability, control, and vendor flexibility over feature rankings, as survey data cited in the source suggests widespread concern about AI vendor lock-in and difficulty switching providers. Voice agents have stricter real-time requirements than text chatbots because transcription, model inference, speech synthesis, turn detection, and interruption handling must operate with low enough latency to feel natural to callers. The proposed evaluation framework focuses on API-level developer control, model and provider choice with bring-your-own keys and fallbacks, pre-launch testing and production observability, lifecycle management from building through optimization, and compliance and scalability for sensitive, high-concurrency use cases. It argues that enterprises can buy a voice orchestration layer rather than build and maintain underlying speech and model infrastructure themselves, while retaining ownership of prompts, workflows, integrations, and model decisions. The source presents Vapi as an API-first, model-agnostic voice platform that provides configurable orchestration, simulation-based evaluations, call monitoring, provider fallbacks, compliance support, and high-volume deployment capabilities, citing customers and usage figures as evidence of enterprise scale.

VapiCon, which overcame initial financial and organizational doubts to draw more than 850 attendees from 451 companies to its inaugural voice AI summit in San Francisco on October 2, 2025, will return as a larger two-day event on November 11–12, 2026, at Fort Mason’s Festival Pavilion. The 2026 conference expects over 1,200 attendees, more than 50 speakers, and over 20 sponsors, reflecting a shift in the voice AI industry from questioning whether the technology works to addressing deployment, continuous agent improvement, and advanced applications. Confirmed sponsors include Deepgram, Krisp, Oxylabs, AssemblyAI, pyannote.ai, Speechmatics, Flexprice, Hamming, and Cekura. Organizers are seeking production voice AI engineers and business decision-makers to submit workshop, presentation, and panel proposals, while tickets are now available and recordings of all 13 sessions from the 2025 event can be viewed in the VapiCon video library.

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.