We initially aimed to teach math but shifted to teaching languages, particularly English, because we realized it would have a broader impact and reach more people.

From the beginning, Duolingo's goal was to teach languages using AI instead of human teachers, as humans are expensive and not scalable.

Our strategy involves expanding our English learning product to include various formats, such as one-on-one and one-to-four teaching models, catering to a wider age range of students from young children to teenagers.

As we branch out into subjects like math and music, the biggest challenge is creating a learning experience that is as engaging and effective as our language courses.

The journey to discovering the potential of language models was not straightforward; we initially explored various other technologies.

We built a culture of iteration and continuous improvement at Duolingo, which has been crucial for our growth.

We always knew we could grow users first, which was our primary focus in the early days of Duolingo.

With AI, we can generate high-quality educational content quickly, allowing us to expand into new subjects like chess, math, and music, enhancing our learning offerings. We started working on the chess course eight months ago, initiated by two people who had no coding or chess experience, and they managed to prototype it successfully before we added engineers to productionize it.

From day one, Duolingo was technology-first, and now we refer to it as AI, which has significantly improved since our inception.