Mati Staniszewski says ElevenLabs started in 2022 after a dubbing problem he saw still unsolved in 2021, when films were dubbed in Polish with a single flat narrator.
ElevenLabs combines frontier audio research with a product platform, and Staniszewski says it passed on lip dubbing and avatar work because video models were not good enough.
Staniszewski predicts voice will become one of the biggest ways people interact with AI within twelve months, a bet that anchors ElevenLabs' strategy.
Why ElevenLabs thinks voice will become AI's main interface
ElevenLabs is betting that voice becomes the main way people talk to AI, and it is building the communication layer for that shift. Its audio research leads today, but Mati expects big labs to narrow the gap. The durable edge therefore comes from products, a creator ecosystem and engineers embedded with customers like Deutsche Telekom.
Takeaways
Voice will become a primary AI interface. Mati predicts voice will be one of the biggest ways people interact with AI within twelve months. Voice carries emotion and imperfection that text lacks, and ElevenLabs' agents performed better once ums and pauses were added. The goal is a Babelfish inside existing devices.
Research edge may shrink, so ecosystem matters. Mati expects the lead over large labs to narrow, so products and a Voices marketplace of 20,000 voices, where creators earn on use, carry more weight. Focus also means pausing weak bets, like lip dubbing two years ago, and never building text-to-video from scratch.
Forward-deployed engineers prove value before scaling. Deutsche Telekom started two years ago with marketing and a daily news podcast, then expanded into call centre agents and a T-Mobile agent that translates calls live, via CRM and phone integrations. FDEs sit in product teams and feed customer lessons back, a Palantir habit.
Flat, transparent organization with small autonomous teams. ElevenLabs runs about five layers at most, managers with close to 10 direct reports, and teams usually under 10 people. Nearly everyone can access company documents. Engineers are embedded in non-technical teams such as legal, and a voice agent talks to website visitors, capturing more than a form.
FinTech adopts conversational agents fastest. Customers such as Revolut and PagBank lead adoption, followed by healthcare and telcos, with retail and e-commerce starting this year. Revenue is strongest in agents and creative work, spread across three core product lines with three more in development.
AI should amplify people, not replace them. Senra criticizes founders who build technology to replace people, and Mati agrees. ElevenLabs has restored voices for over 10,000 people who lost them to illnesses like ALS and cancer, working with 800 partner organizations, including Emmy winner Tim Green, who now hosts a podcast.
Turning down acquisitions to go all in. ElevenLabs has received three or four concrete offers, the latest last June. Mati, 31, chose to keep building, saying the biggest risk is taking none. Senra argues money cannot replace work founders love, citing a hypothetical $2tn offer to Steve Jobs.
How a dubbing problem became a voice research company
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Mati and Piotr Dąbkowski, best friends for fifteen years, did hack weekend projects while at Palantir and Google, then founded ElevenLabs in 2022.
Films in Poland were still dubbed by a single flat narrator in 2021, stripping out the original emotion.
Dubbing needed several steps from transcription through speech regeneration, and the research behind each was weak.
Creators wanted narration tools first, so the team tackled speech generation and then product delivery, with Piotr assembling the research team.