Descriptions:
Luke Harries of ElevenLabs presents an early preview of Voice Engine at the AI Engineer conference — a new product designed to let developers add voice to existing chat agents without rebuilding their stack from scratch. The motivation: most companies have already invested heavily in chat agent infrastructure (evals, RAG pipelines, tool calling), and a full rebuild to add voice introduces unnecessary risk and rework.
Voice Engine wraps any existing agent as a first-class primitive, handling the complete voice stack: speech-to-text via Scribe (ElevenLabs’ most accurate model), text-to-speech via V3, and advanced turn-taking that is emotion-context aware and handles semantic pauses. The server SDK requires minimal new code — a client instantiation, a voice engine wrapper, and a session loop that proxies traffic to the underlying chat agent. A three-line client SDK adds an embeddable widget to any website, and telephony and SaaS channel support come automatically once the client SDK is in place. UI components are built on shadcn/ui and Vercel conventions, making them drop-in compatible with modern frontends.
Harries demonstrates the upgrade path live: a generic chat support agent running locally is converted to a voice agent using a single Claude Code prompt, which analyzes the codebase, identifies the existing agent structure, and writes the wrapper automatically. ElevenLabs plans to ship a Claude skill bundling all recommended configuration for the pattern. The product was expected to launch publicly within weeks of the AI Engineer conference.
📺 Source: AI Engineer · Published May 09, 2026
🏷️ Format: Keynote Launch
