How Speak Built a $1B AI Language Learning App

14 snips

Feb 5, 2026

Connor Zwick, cofounder and CEO of Speak, built a speech-first, AI-powered language app that reached a $1B valuation. He discusses leaving Harvard for deep learning, building superior speech models with Andrew Hsu, launching in Korea to rigorously validate product-market fit, and scaling to multiple languages while solving reliability, moderation, and growth challenges.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

INSIGHT

Speaking Is The Missing Core

Traditional language apps teach vocabulary but fail to get learners speaking aloud enough to build fluency.
Speak centers on voice-first practice because speaking without fear of judgment accelerates progress.

ADVICE

Use A Technical Wedge To Build A Flywheel

Start with a technical wedge that you can out-execute incumbents on and use it to build data and scale.
Iterate models, gather usage data, and expand the product flywheel from there.

ANECDOTE

Users Speak Freely To AI

Early user tests showed people edited themselves when practicing with others but spoke freely with an AI alone.
That lack of fear produced surprisingly positive engagement and enjoyment in trials.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

Connor Zwick is the cofounder and CEO of Speak, a billion dollar app that is reinventing digital language learning by getting users into real conversations with AI-powered voices. Connor started Speak in 2016 with cofounder Andrew Hsu. Prior to founding Speak, Connor built his first startup called Flashcards+ when he was still in high school and eventually sold it to Chegg. He then went to Harvard, but dropped out to become a Thiel Fellow and found another startup called Coco Controller.

Key Takeaways:

After dropping out of Harvard to pursue a Thiel Fellowship, Connor became very interested in deep learning along with another Thiel fellow named Andrew Hsu.
After Connor and Andrew developed a speech recognition model that outperformed all of the existing models at the time, they realized that AI had the potential to revolutionize language learning, which led them to cofound Speak.
From the beginning, Connor and Andrew believed there was a major gap in the language learning market around immersive language learning. Unlike Duolingo and Rosetta Stone, which focus on helping people learn vocabulary, Speak went all in on AI-powered voice right away to help students learn by speaking.
For the first four years, Speak exclusively focused on English language learning in South Korea. This is a ruthlessly competitive market, with South Koreans spending 1% of GDP learning English, but this forced Speak to tune its model to be great at one language before the company decided it was ready to expand.
Since international expansion was part of Speak’s DNA from day one, it allowed the company to expand quickly to other markets once it was ready to grow beyond South Korea. Today, Speak helps people around the world learn six languages, including English, Spanish, French, Italian, Japanese, and Korean. Based on this growth, the company recently raised a Series C at a $1B valuation.

Connor Zwick:

Website: https://www.speak.com/
LinkedIn: https://www.linkedin.com/in/connorzwick/
X: https://x.com/connorzwick

Phil Carter:

Website: https://www.philgcarter.com
Substack: https://philgcarter.substack.com
LinkedIn: https://www.linkedin.com/in/philgcarter
X: https://x.com/philgcarter