ElevenLabs CEO: Why Voice is the Next AI Interface

How ElevenLabs keeps shipping fast at 350 people

ElevenLabs keeps shipping quickly at 350 people by organising around small teams that must prove themselves and by letting research solve product problems. Staniszewski argues the hard part of scaling sits in the transitions, from a Europe-born remote startup into enterprise sales and licensed music. In each, hiring, incentives and trust with incumbents trail the strategy and need deliberate fixes.

Takeaways

• Small, disposable teams drive ElevenLabs' shipping speed. About 20 product teams of five to ten people give high ownership, at the cost of some duplicated work. Job titles were removed a year ago. Each new team gets six months to prove itself or is shut down.
• Research solves problems instead of interface knobs. ElevenLabs resisted requests for a voice speed slider for about nine months to avoid becoming a previous-generation editing suite, then solved speed control in research. When research needs more than three months, product teams build other models meanwhile.
• Starting in Europe was a structural advantage. Staniszewski says ElevenLabs would not exist otherwise. Polish dubbing often uses one monotone narrator. It started fully remote, hired unconventional talent like a call center worker who built an open source text to speech model, and opened hubs past 30 people.
• Creative partners need incentives understood and technology shown. The Voice Marketplace has almost 10,000 voices in 70 languages and has paid creators $10m. The licensed music deal with Merlin and Kobalt took eighteen months, helped by forcing functions and demos to label members that defused reflexive rejection.
• The enterprise gap is demo to production. Starting with Hippocratic AI's appointment calls in 2023, ElevenLabs chose to handle full orchestration across speech to text, language models and text to speech. Telephony integration such as Twilio, testing, monitoring and four or five nines of reliability remain hardest.
• Sales incentives lag strategy and need overriding. Having engineers sell failed, so the mix is now 80% sales and 20% engineering. Reps flag deals that look competitive but wrong and still get commission when they are killed. Sales to foundational model companies are barred after a competitor sought licences.

Playbook for entering new domains

• Spend time understanding the priorities and incentives of the industry before pitching it.
• Hire at least one or two people who already know the space, and add outside consultants, as ElevenLabs did with music lawyers.
• Look for a legal counsel who acts as a thought partner. A hire from several Fortune 500 companies with no startup experience mostly flagged risks.