Jun 20, 2025 · 1h 19m · 20product
20Product: How Duolingo Build Product 10x Faster with AI | Duolingo's Biggest Lessons on Paywalls, Push Notifications and In-App Purchases | Why Small Teams are the Future of Product | Why PMs Will Become Extinct with Cem Kansu, CPO @ Duolingo
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of 20 Product, Duolingo CPO Cem Kansu reveals how the world's leading education app leverages AI prototyping to build products 10x faster while optimizing freemium monetization, gamified retention mechanics, and brand expansion.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 28% of the talking time here. How this is scored →
Harry as informed peer 2.8 Guest teaching 4.0 Guest disagreement 1.4 Harry pushing back 2.0
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 22:40 Dismissing AI productivity hype as autocomplete
Cem directly counters popular industry stats regarding AI code generation from major tech players, characterizing current gains as basic autocomplete functions rather than true productivity multipliers.
Hardest push from Harry ▶ 22:20 Harry pushes back on minimal AI engineer efficiency gainsHarry refuses to accept Cem's estimate of modest 10-20% developer productivity gains, citing conflicting claims from companies like HubSpot, Salesforce, and Microsoft.
Biggest teaching moment ▶ 40:40 Conversion psychology of explicit zero pricing stringsCem educates Harry on micro-copy optimizations, demonstrating how formatting price trial buttons with explicit decimal zeros consistently outperforms standard free trial copy.
Harry holds his own ▶ 6:05 Citing Gustav Soderstrom on product detailsHarry demonstrates high product craft fluency by invoking Spotify CPO Gustav Soderstrom's thesis that fine details constitute the actual product itself.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Duolingo's Product Philosophy and Attention to Detail | 4 | 3 | 1 | 1 | Harry opens by quoting Spotify CPO Gustav Soderstrom on how product details are the product itself. Cem agrees and forcefully reframes the common trope of PMs being 'mini CEOs' as the worst definition in product management. | |
| Design Evolution and AI-Powered Video Call Prototyping | 2 | 3 | 0 | 1 | Harry asks detailed questions about Figma usage and rapid prototyping workflows. Cem breaks down Duolingo's iterative review process using the AI Video Call with Lily as a primary case study. | |
| Bottoms-Up Innovation and Building Products 10x Faster | 3 | 3 | 0 | 2 | Harry quotes Matthew McConaughey regarding constraints and options to query feature creep versus depth. Cem details how two employees vibe-coded the new chess course without engineering staff. | |
| Transforming Product Workflows and AI Coding Tools | 5 | 4 | 3 | 5 | Harry challenges low 10-20% engineer productivity gain claims by citing AI code stats from Salesforce and Microsoft. Cem dismisses these stats as mere autocomplete gains, leading to an active debate on vendor defensibility. | |
| Duolingo's Monetization Pivot and Paywall Optimizations | 3 | 5 | 2 | 3 | Cem gives a detailed masterclass on paywall psychological levers, explaining why adding multiple zeros like 0.00 dramatically boosts conversion. Harry questions pricing tier structures and choice paradoxes. | |
| Revenue Streams, Brand Expansion, and Screen Time Competition | 2 | 4 | 3 | 1 | Cem explicitly rejects categorizing Duolingo as traditional edtech or gaming, criticizing edtech products as boring ass products and clarifying that Duolingo competes strictly for screen time. | |
| In-App Purchases, Virtual Economy, and Habit Building | 2 | 3 | 1 | 1 | Cem explains the virtual gem economy and streak freeze mechanics. He points out that monetizing core lesson progress directly is a short-sighted strategy that destroys user trust. | |
| Retention Strategies: Streaks, Leaderboards, and Push Notifications | 2 | 6 | 1 | 2 | Cem outlines the three primary drivers of Duolingo retention: streaks, leaderboards, and push notifications. He shares a specific data insight regarding scheduling notifications precisely 23.5 hours after a user's previous session. | |
| Product Failures, Unexplored Ideas, and Social Strategy | 2 | 5 | 2 | 2 | Cem analyzes past product missteps including teacher dashboards and an Uber-style live human tutoring feature that failed because users felt intense social anxiety when stumbling on video calls. |
Statements from this episode (28)
Kansu: Duolingo enforces specific design language and pixel-perfect details on every screen
“No, we take pride a lot in the pixel perfect details. Like, I think you'll see that in the product, right? It's like every screen will have a very specific design language. Characters will be very specific, etc.”
Kansu: Consumer products live and die in the pixels
“Consumer products live and die in the pixels. Meaning if you are working on a consumer product, you have to figure out how to make your change pixel perfect for the user.”
Kansu: Calling a product manager a "mini CEO" is the worst definition
“Product managers are sometimes viewed as these like, I'm the mini CEO. I decide strategy and then people execute. That is like the worst definition of a product manager that I've ever heard.”
Kansu: Duolingo Math and Music took double the time of AI-prototyped Chess
“Our experience with doing this is, was with math and music before. It was triple the amount of people and double the amount of time it takes to get a course out.”
Kansu: Duolingo aims to improve existing subjects 10x rather than expand subject count
“The goal is not to get to 40 subjects and do them all equally the same. I think make what we have, Tenex, better is the primary goal.”
Kansu: AI tools will cause engineering, product, and design roles to converge
“I think what is likely to happen is there is clear work definitions of what an engineer does today, what a product manager does, and what a designer does. I think these are going to converge way more than what's happening today in technology organizations.”
Kansu: AI product tools will generate complete apps within 12 months
“In my opinion, that will get solved in, I don't know, maybe 12 months, but I think if you know how to use these tools to create product, build product, and you can kind of harness their power to build great things, you'll come out ahead.”
Kansu: Traditional product specs and linear PRD processes are dead
“Specs, like, you know, companies write product specs that like go from someone's idea that turns into a product spec, and then that goes into design, that goes into engineering. This process is like dead. Like, why would you write specs if you can go from your…”
Kansu: Current AI coding tools only boost developer productivity by 10-20%
“I think my read right now is more of the latter. And also the magnitude of multiplication is way lower than like 10 X-ing things. I think it's maybe in the realm of like 10 to 20% more productivity, Of an average engineering capacity.”
Kansu: Cursor sees higher internal adoption and output than Windsurf at Duolingo
“Because internally at Duolingo, I'm seeing way more usage and way more creation come out of Cursor.”
Kansu: Tasteful ad integration does not hurt consumer app retention
“Putting ads on Duolingo, if you do it tastefully, doesn't kill your retention.”
Kansu: Compromising free user experience for ad revenue destroys organic growth
“The moment you ruin your free user experience, you kill your growth engine.”
Kansu: Freemium models create more defensible consumer app businesses than premium-only
“If you want to build a large business, it's way more defensible to build a great free user experience and do it freemium rather than full premium.”
Kansu: Paywall buttons displaying '$0.00' significantly boost trial subscription conversion
“If you write multiple zeros, and you hint at the fact that there's a free trial with multiple zeros, conversion goes up a significant amount.”
Kansu: Two to three subscription choices is the conversion sweet spot
“If we give you more than three packages on one page, it's too much, like too much choice. You also convert way less. I think two to three is, we've seen as a sweet spot.”
Kansu: Over 50% of subscribers choose Duolingo's recommended middle option
“I don't remember the exact percentage, but certainly more than, it's the majority, more than 50%.”
Kansu: Merchandising will become a much larger revenue line for Duolingo
“We do have this revenue line, but it's incredibly small today, but I think it will be way bigger is merchandising.”
Kansu: Products in the edtech space are all boring
“The reason we don't fit into education very much is if you look at the education space, honestly, this is a bit of a hot take, but it's all boring ass products.”
Kansu: Duolingo competes with social media and gaming for screen time
“Everything that takes up your screen time, that might be social media, that might be gaming, that might be streaming, whatever it is, at the end is what Duolingo competes for.”
Kansu: Daily retention is the king of consumer product metrics
“Daily retention is the king of all metrics for consumer products.”
Kansu: Users reaching a 30-day Duolingo streak have near 100% retention
“What we intend to see is if you've reached a 30 day streak, so you've built up a 30 day streak, your retention is very close to a hundred percent.”
Kansu: Scheduling push notifications at 23.5 hours drives highest user retention
“We schedule a notification Almost exactly at 23 and a half hours later from that time, and we've tested many different timing, and the notification will remind you to practice your Spanish, because we, you just did a Spanish lesson. We've tested many different…”
Kansu: Duolingo previously tested an on-demand live human tutoring subscription
“There was a short lived moment where we had a high tier subscription, which was basically an Uber for a Spanish tutor. So you could press a button to speak Spanish to a Spanish speaker. Immediately.”
Kansu: AI conversation features outperform human tutoring by eliminating user embarrassment
“AI is way better for this, because you can stumble, you can talk shitty Spanish, you're not embarrassed. A lot of users found it embarrassing to stumble when there's a human looking at them on video.”
Kansu: Languages remain vast majority of Duolingo usage, math and music small
“Majority of our users are doing languages. Chess is very new, like counted as kind of zero still. It just rolled out like a few days ago. So, but math and music are a small percent of our business.”
Kansu: AI will completely transform product management within two years
“Honestly, now, probably in 12 to 24 months, the job of a PM will look entirely different. So I think we're talking about five years because that feels safe, but honestly, my guess now, or what I've changed my mind on is, it's going to be in two years, is my gu…”