May 2026 Summaries

Netflix and Doritos collaborated to create a unique promotional campaign for the final season of Stranger Things, where fans could engage in an interactive experience reminiscent of an '80s telethon. This involved calling a number on special-edition Doritos bags to interact with AI-powered versions of '80s icons like David Hasselhoff and Paula Abdul, in a choose-your-own-adventure format. The system used a hybrid approach, integrating voice AI with a library of pre-recorded celebrity voiceovers, to maintain a human touch as desired by the show's creators, the Duffer Brothers. Casper Studios chose LiveKit Cloud to manage the complex task of handling live sessions with dynamic, non-linear call flows, allowing for scalable, real-time interaction without infrastructure management. This innovative approach resulted in over 400,000 interactive call sessions, highlighting a potential blueprint for future voice-driven marketing campaigns across various brands, as Omnicom and partners explore similar experiences.

Telli, a company founded in 2024 by Finn zur Mühlen, Philipp Baumanns, and Seb Hapte-Selassie, automates high-volume phone operations for B2C enterprises using AI voice agents to handle tasks like lead qualification, appointment bookings, and customer service. The company quickly gained traction, securing a $3.6 million pre-seed investment from Y Combinator and scaling its revenue significantly. Telli leverages LiveKit for its unified platform, which supports reliable telephony, real-time agent runtime, and compliant call recording, enabling seamless integration with various telecommunication infrastructures. The switch to LiveKit allowed telli to rapidly migrate and operate across multiple languages and regions while ensuring low latency and high audio quality, crucial for maintaining customer engagement and satisfaction. Telli also collaborates with ai-coustics to enhance audio quality, addressing challenges such as short utterance transcription and turn-taking errors, which are critical for enterprise-level deployment. This partnership has helped telli expand its services across Europe, working with major enterprises like Sky to improve customer engagement and operational efficiency, ultimately leading to significant revenue growth and establishing telli as a trusted provider in the voice AI market.

LiveKit, an open-source project used for real-time audio and video applications, was recently scrutinized following external claims that audio quality degrades when video is also shared. Contrary to these reports, tests conducted with LiveKit's architecture demonstrated that audio quality remains unaffected by adding video tracks, as audio and video are carried on separate RTP streams, preventing interference. The tests involved three scenarios: audio only, audio with a 720p webcam, and audio with webcam and screen sharing, all of which showed consistent audio quality. Results indicated that LiveKit's adaptive stream and dynacast settings efficiently managed bandwidth without degrading audio quality. While acknowledging the limitations of these tests, such as the lack of geographically distributed participants and varied network conditions, the findings suggest that LiveKit effectively isolates audio from potential video interference, maintaining stable audio performance even when video demands change.

LiveKit offers a simplified solution for embedding voice agents into websites by using a single <script> tag, eliminating the need for frontend integration, SDKs, or custom call UIs. This feature allows websites to provide real-time voice interactions akin to an Intercom chat bubble, but through voice, enhancing user experience by engaging them in context-specific conversations directly on the page they are viewing. The voice agents can be used for various applications, such as e-commerce assistance, live support and onboarding, lead capture, and even after-hours inquiries, all while maintaining the user's flow. The setup involves generating a snippet within the LiveKit Cloud dashboard that can be easily embedded into platforms like WordPress, Shopify, or custom sites, and it supports additional features like video, screen sharing, and text chat. Customization options are available for branding, user context, and domain lock settings, ensuring the widget fits seamlessly into the site’s aesthetics and maintains security by only loading on approved domains. The embedded sessions integrate with LiveKit's Agent Console for comprehensive management and analytics.

Answering Machine Detection (AMD) has been integrated into LiveKit Agents, enhancing their ability to identify whether an outbound call reaches a human, voicemail, IVR, or is unavailable, without requiring additional plugins. This technology allows agents to quickly determine the nature of the call within the first second, enabling appropriate responses such as leaving a message, navigating menus, or ending the call, which improves efficiency and reduces costs. AMD operates outside the agent's main loop, using a short-circuiting rule and LLM classification to ensure rapid and accurate detection, with a strong reliance on transcripts from the STT pipeline. The system is highly configurable, allowing for various levels of conservatism depending on the use case, and supports auto-IVR navigation and interruption protection to prevent unnecessary resource use. Benchmarked against diverse datasets, the best-performing model pairing achieves high accuracy and quick decision times, with options to swap models depending on user preference. AMD is now available for Python and Node.js, with extensive documentation for configuration and integration, aiming to streamline the workflow of outbound call agents.

SAP is revolutionizing enterprise software by incorporating AI, specifically through its AI offering, Joule, which is embedded across its cloud portfolio like SAP S/4HANA and SAP SuccessFactors. Joule enhances productivity by up to 75% by delivering contextual insights and automated actions within familiar workflows. Traditionally reliant on text-based interactions, Joule's capabilities are being expanded with LiveKit's real-time voice technology, allowing for natural language engagement in a variety of settings beyond the keyboard. This voice integration extends Joule's utility to diverse roles and environments, such as warehouse floors and field operations, making AI a more integrated and intuitive part of the enterprise experience. LiveKit provides the technological backbone for Joule's voice features, offering an enterprise-ready platform designed for natural, context-aware conversations and enabling SAP to maintain control over its AI's reasoning and business logic. The voice integration is currently available through SAP's Early Adopter Care program, with a general release anticipated in the second half of 2026.

Voice agents are increasingly prevalent in various customer service roles, and users are becoming adept at identifying generic text-to-speech (TTS) voices, prompting a need for customized voice solutions to better align with brand identity and specific use cases. LiveKit Inference addresses this by enabling the creation of cloned voices that can be used across multiple TTS providers, providing consistency and resilience in case of service failures. By uploading or recording an audio sample, users can generate a single voice ID that links to multiple providers, allowing for flexibility in selecting the optimal voice model while maintaining continuity with automatic fallback options. This feature, now available on LiveKit Cloud, streamlines the integration of customized voices into voice agents, with no additional cloning fees, enhancing both user experience and operational efficiency.