Parrot Thinks Duolingo Got Language Learning Wrong, and 6-Hour Power Users Suggest They Might Be Right

The Macro: Nobody Actually Learns a Language on Duolingo

I am going to say something that the language learning industry already knows but rarely admits out loud: the dominant apps do not produce fluent speakers. Duolingo has over 100 million monthly active users and a $7 billion market cap. It is an incredible business. It is also, by the admission of most serious linguists, not a great way to actually learn a language. The gamification loop is optimized for retention, not acquisition. You keep your streak alive. You feel productive. You learn to translate disconnected sentences about red cats and tall buildings. Three years later, you land in Tokyo and cannot order dinner.

This is not a controversial take among language acquisition researchers. The dominant methodology in the field is called Comprehensible Input, popularized by Stephen Krashen decades ago. The idea is simple: you acquire language by being exposed to content you can mostly understand, with enough new material mixed in to stretch your abilities. Not flashcards. Not grammar drills. Not gamified translation exercises. Actual content in the target language that you find interesting enough to keep consuming.

The problem has always been delivery. Comprehensible Input works, but curating the right content at the right difficulty level for each learner is labor-intensive. Teachers can do it in classrooms. Apps have struggled to automate it. Until now, because AI changes the content curation problem completely.

The language learning app market is crowded. Duolingo, Babbel, Busuu, Rosetta Stone, Pimsleur, Drops, LingQ, Lingvist, HelloTalk, Tandem, italki. I could keep going. The space has more apps than most learners have attention spans. But almost all of them are built on the same pedagogical model: structured lessons with progressive difficulty. The feed-based, video-first approach is genuinely different.

The Micro: 52 Apps Tested, Zero Satisfaction

Parrot was founded by Amir P. Hanna (CEO), Erik Dahl (CTO), and Julia Hudea (CPO). They came through Y Combinator’s Fall 2025 batch with what they call “TikTok for language learning.” Before writing a single line of code, they tested 52 language learning apps and interviewed 65 learners. That is an unusual amount of user research for a seed-stage startup, and it shows in how specifically they have identified the gap.

The product works like a personalized short-form video feed. Instead of swiping through dance videos and cooking clips, you are scrolling through content in your target language, calibrated to your comprehension level. The AI personalizes the feed so that each video sits in that sweet spot of being mostly understandable but challenging enough to push your skills forward. This is Comprehensible Input delivered through the content format that has proven most engaging for modern attention spans.

The engagement numbers they are reporting are striking. Power users spend roughly 6 hours per week on Parrot. For context, Duolingo’s average session length is about 7 minutes, and most users open the app once a day to maintain their streak. Six hours a week is not streak maintenance. That is genuine immersion behavior. If those numbers are accurate and sustained, Parrot has found something that the gamification-first approach fundamentally cannot deliver: users who actually want to spend time with the content.

Early validation was also promising. When they showed their MVP to surveyed students, 60% expressed willingness to pay $30 for early access. In consumer apps, stated willingness to pay is always higher than actual conversion, but 60% is a strong signal that the value proposition resonates.

The team is three people in San Francisco. Amir runs the business, Erik handles the technical architecture, and Julia drives the product experience. For a consumer app that depends on content quality and personalization, having a dedicated CPO from day one is smart. A lot of language apps feel like engineering projects with content bolted on. Having someone focused on the learning experience as a first-class concern should prevent that.

What competitors should worry about is the TikTok comparison. Not because Parrot is going to become a social media platform, but because the algorithmic feed model has proven it can hold attention at scale. If Parrot can combine that engagement mechanic with actual pedagogical value, the resulting retention curve could look very different from traditional language apps.

The Verdict

Parrot is making a bet that the feed format will beat the lesson format for language acquisition. I think the bet is directionally correct. The research supports Comprehensible Input. The engagement data supports feed-based consumption. The combination has not been executed well before, partly because personalizing video content at the right difficulty level required AI capabilities that did not exist two years ago.

The risk is content. A video feed needs an enormous volume of high-quality content to feel fresh. If users see the same clips recycled, engagement will crater fast. Building or sourcing that content library is expensive and operationally complex. Duolingo has thousands of contributors and years of content accumulation. Parrot is starting from zero.

The other risk is the classic consumer app problem: attention is zero-sum. Parrot is not just competing with other language apps. It is competing with TikTok, Instagram Reels, YouTube Shorts, and every other thing a person could do with their phone for six hours a week. Positioning as “TikTok for language learning” is a smart framing for fundraising, but in practice, you are asking users to choose educational content over entertainment content in the same format. That is a hard sell at scale.

At thirty days, I want to see week-two retention. The 6-hour power user stat is meaningless if those users churn after a month. At sixty days, I want to see content pipeline velocity: how fast can they produce or curate new material? At ninety days, the question is monetization. Can they convert free users to paid at a rate that supports the content production costs? The product insight is genuine. The research homework is impressive. Now they need to prove that Comprehensible Input delivered through a feed can scale beyond early adopters into a mainstream learning tool.