Key Takeaways

  • For decades, humans bent their behavior around machines by learning keyboards, screen layouts, and programming languages. Voice models reverse that relationship.
  • ElevenLabs co-founder Mati Staniszewski built the speech AI company after growing up in Poland frustrated by poor movie dubbing where a single voice read every line.
  • Hardware gadgets are a distraction; the future interface is software-driven voice embedded inside existing phones, laptops, and earbuds.
  • Staniszewski structures ElevenLabs with small, flat teams and forward deployed engineers, drawing directly from his operating experience at Palantir.
  • True conversational AI requires pairing raw intelligence (IQ) with emotional context (EQ) so models can pause, think, and reinsert themselves naturally.

The Machine Must Learn Human, Not the Reverse

Every major computing era forced people to master unnatural habits. You had to type on a QWERTY keyboard, swipe across glass, or write code to make a computer do what you wanted.

Staniszewski sees that dynamic ending. As raw machine intelligence grows, the interface layer must return to humanity's oldest communication tool: voice.

“For decades, we learn how technology around works and learn the language of that technology, the keyboard, the screen, the coding languages, even,” Staniszewski explained. “You need to learn how the technology works so you can control it. And now, what I think we can solve is flip it back to how we want to communicate, how the most primal way is voice conversation, and you can bring technology on our terms.”

Instead of building another piece of plastic you carry in your pocket or clip to your shirt, Staniszewski wants current screens to fade into the background. “Hopefully, the screen, the phone will kind of be able to be back in the back pocket if you won't need it in the same way as you do now.”

EQ Is the Missing Half of Intelligence

Most foundation model companies chase reasoning benchmarks and context windows. That gives a model IQ, but raw logic falls apart in real-time voice interactions if the system lacks conversational timing.

A natural voice interface cannot just read text fast. It must read tone, sense hesitation, manage interruptions, and match the emotional state of the speaker.

“Voice is definitely part of it, but it kind of understands you,” said Staniszewski. “It's able to connect the IQ plus the EQ part of it. Emotionally understand and know how you're feeling, can adjust based on that, can pause, can think, can reinsert itself into the conversation.”

This principle guided Staniszewski to reject acquisition offers and keep ElevenLabs solely focused on speech and audio research. Instead of trying to build an all-in-one AI ecosystem, the company focuses on creating the voice infrastructure that every other application plugs into. The ultimate goal is a universal translation layer embedded everywhere: “Yes, Babel Fish will exist, and we'll hopefully make it happen, but less so that we'll create Babel Fish itself, but we will enable everyone out there to have Babel Fish in their existing devices, existing presence, existing work.”

What to Do With This

Audit your product's user friction by listing every input field or click path required to get an answer. Test a prototype where a user speaks their request in one sentence and receives an audio response, skipping the visual UI entirely. If your team cannot ship a voice test in two weeks, assign one forward deployed engineer to build directly with ElevenLabs or OpenAI real-time audio APIs.