No Priors Ep. 143 | With ElevenLabs Co-Founder Mati Staniszewski

ElevenLabs structures work into ‘labs’—small cross‑functional teams that first solve a core technical problem (e.g., voice, agents, music) and then quickly ship product layers on top, so research advances are immediately usable by customers.

Staniszewski argues that keyboards and screens are anachronistic for many tasks; natural speech will increasingly mediate interactions with phones, computers, robots, and services, requiring high‑quality, expressive, low‑latency audio models.

Enterprises often don’t know how to ‘choose’ a voice, so ElevenLabs supports them with internal ‘voice sommeliers’ and dynamic voice routing based on user segments, proving that tone, accent, and emotion heavily drive user perception.

Customers like Meesho use agents not just for refunds and tracking, but as front‑door shopping assistants that help discover products and guide checkout, pointing to a shift from narrow support bots to full journey companions.

ElevenLabs expects base models to commoditize within a few years; they view research as a 6–12 month advantage that lets them build better products and, more importantly, an ecosystem of voices, integrations, and workflows that compound over time.

Despite competition from OpenAI and Google, ElevenLabs claims to outperform them on key audio benchmarks by focusing on architectural breakthroughs rather than sheer scale, and by concentrating some of the world’s top audio researchers under one roof.

From chess.com tutors in the voices of top grandmasters to Ukraine’s push toward an ‘agentic government’ with voice‑driven citizen services, the same audio and agent infrastructure can reshape how people learn, access benefits, and interact with institutions.