Inside Duolingo’s AI Video Call: A Revolution in Language Learning

Feb 12, 2025

👋 Welcome to the latest issue of the AI Community Learning Series! We just wrapped up an insightful session with Zan Gilani, Principal Product Manager at Duolingo, where we explored how AI-powered Video Call is transforming language learning. If you missed it, here’s a recap of the key takeaways from our conversation and our session recording.

AI-powered conversations make language learning more engaging and stress-free.

Many learners struggle with real-world conversations, often freezing when speaking with native speakers. Duolingo’s AI-powered Video Call addresses this by offering a stress-free, engaging, and adaptive speaking experience.

“Conversation is a muscle. The only way to get better is to practice.” – Zan Gilani

The Thoughtful Design Behind Lily

When Duolingo set out to create an AI-powered conversation partner, the goal was clear: make AI interactions feel human and engaging, not robotic and repetitive. Enter Lily, a virtual character with personality, wit, and a uniquely sassy attitude.

The team could have easily built a generic AI assistant that responded predictably and politely. But they realized that for users to stay engaged, Lily had to feel like a real conversation partner—one who could surprise, challenge, and amuse. This wasn’t just about fun; it was about replicating the unpredictability of real conversations.

The challenge, however, was striking the right balance between personality and usability. If Lily was too sarcastic, she could demotivate learners. If she was too neutral, she would blend into the background like any other AI chatbot. Through continuous iteration and user testing, the team fine-tuned her responses, making her slightly unpredictable but still supportive.

Not Just Another Chatbot: Lily is designed to be a realistic language partner, not just a scripted AI assistant.
Adaptive Learning: The AI adjusts its conversation style to match skill levels, from structured beginner dialogues to open-ended discussions for advanced learners.
Reducing Cognitive Load: Duolingo’s AI simplifies the learning process, allowing users to focus on practicing rather than structuring their own lessons.

Striking the Balance Between Engagement and Learning

Designing AI for education is not just about maximizing what it can do—it’s about ensuring it serves a purpose. The Duolingo team learned that too much AI capability can be overwhelming, while too little can feel unhelpful. Their solution? Intentional constraints that enhance learning rather than distract from it.

The key to this balance was understanding user behavior. Early testing showed that if conversations ran indefinitely, learners often disengaged. Instead of making AI endlessly responsive, Duolingo ensured that Lily always ends the call—a simple, yet powerful way to leave learners eager for the next session. Similarly, while AI can generate infinite variations of responses, Duolingo deliberately limits interactions to maintain focus and effectiveness.

This approach ensures that AI remains a tool for engagement, not an endless distraction. With this mindset, Duolingo crafted an AI-powered experience that is dynamic yet structured, playful yet purposeful.

From Experimentation to Innovation: How Duolingo Perfected AI Learning

The journey to building Duolingo’s AI-powered Video Call wasn’t a straight path. The team started with a broad range of AI-driven learning features—voice-based lessons, real-time AI tutors, and interactive exercises. But through months of rigorous testing, they kept returning to one crucial insight: real-time speaking practice had the highest impact on language retention and confidence.

With this realization, the team embarked on a series of refinements. Early versions of Lily leaned heavily on humor and sarcasm, but user feedback showed that too much of it was demotivating. The team responded by fine-tuning Lily’s personality—keeping her playful but ensuring that she remained an encouraging learning partner.

Balancing engagement and learning also required striking the right emotional tone. Too much AI-driven correction risked making learners feel self-conscious, while too little feedback left them unsure of their progress. Duolingo chose to focus on encouraging learners through conversational flow, rather than interrupting them with corrections, helping users build confidence in their speaking skills.

Each iteration was driven by data and real-world feedback. With every tweak, the AI evolved into a more effective and engaging learning tool. The result? A uniquely human-like AI experience that motivates users to return and practice consistently.

Reflections from Zan: Lessons from a Decade at Duolingo

Having spent nearly a decade at Duolingo, Zan Gilani has seen firsthand how the company’s philosophy of simplicity and rapid experimentation shapes its success. Reflecting on his journey, he shared two critical insights:

Simplicity and speed are core values. In product design, it’s easy to overthink solutions. Instead, Duolingo emphasizes decisive action—testing ideas quickly rather than waiting for the perfect answer.
The real skill is making complex problems simple. Great product managers don’t just create sophisticated solutions; they distill challenges into clear, manageable steps that drive progress.

This mindset of continuous learning and iteration underpins Duolingo’s ability to stay at the forefront of AI-driven education, ensuring that each feature—like AI-powered Video Call—is optimized for engagement, usability, and real-world impact.

👉 Read more directly from Zan on How experimentation builds breakout features at Duolingo

Learn AI the Fun Way?

Try our AI Trivia Game and join the AI Enthusiast Community on Microsoft Teams! Sign up with your personal email (not your work email) to connect with like-minded people, explore AI insights, and discover exciting projects! 🚀

Let's Play!