The Future of Voice AI: Agents, Dubbing, and Real-Time Translation with ElevenLabs Co-Founder Mati Staniszewski

What It Covers

ElevenLabs co-founder Mati Staniszewski discusses building a voice AI company from zero to $300M ARR in three years, spanning creative dubbing tools and conversational agent platforms serving 5M monthly users across enterprise and self-serve segments.

Key Questions Answered

• •Multi-segment business model: ElevenLabs splits revenue 50-50 between self-serve creator subscriptions and enterprise sales-led agents platform, serving both individual audiobook narrators and Fortune 500 customer support systems with the same foundational voice models underneath both product lines.
• •Voice quality differentiation: Audio AI requires architectural breakthroughs over compute scale. ElevenLabs employs roughly 10 of the estimated 50-100 top audio researchers globally, focusing on controllability and emotional expression rather than just transcription accuracy to beat larger labs on benchmarks.
• •Research-product sequencing: Teams get three months to ship product solutions before waiting for research breakthroughs. This parallel approach prevents product roadmaps from stalling on uncertain research timelines while allowing engineering to build integrations and workflows that capture value when models improve.
• •Cascaded versus fused architectures: Enterprise conversational agents require cascaded speech-to-text plus text-to-speech systems for reliability and tool-calling capabilities today. Fused speech-to-speech models enable more expressive consumer experiences but lack the structured control enterprises need for production deployment over the next year.