When Unhinged Becomes a Product Philosophy
Most companies treat the word "unhinged" as a warning sign. At Duolingo, it's the ultimate badge of honor. When Cem Kansu and his team review a product change and declare it "unhinged," they're not issuing a critique—they're celebrating. The metric isn't whether a feature feels polished or safe; it's whether it breaks the sanitized norms of consumer product design.
This isn't aesthetic rebellion. It's strategic character-building. Duolingo's app icon has appeared with wrinkles, a melting face, and boogers dripping from Duo's nose. Each time, the change ran for two to three weeks. The trigger was precise: users who'd lapsed felt personally haunted. They believed Duo was reacting to their guilt, even though the icon changed globally. The personalization was an illusion, but the emotional punch was real—and retention spiked.
We initially designed Duo to be a motivating coach within the app, but social media made it like this haunting person that if you don't do your lessons, he will haunt you down. That wasn't necessarily our creation, but we certainly over time leaned into it.
— Cem Kansu
When Kansu reviewed the "sick Duo" icon, his feedback wasn't about softening the design. It was that there wasn't enough booger. The centralized product review process—where CEO Luis von Ahn, Kansu, and the head of design participate—doesn't gatekeep weirdness. It enforces it. The result is a product that doesn't just teach Spanish; it haunts you into fluency.
Consumer Products Live and Die in the Pixels
Kansu rejects the "mini-CEO" framing of product management. It's not just wrong—it's the worst definition he's ever heard. The CEO is the CEO of the product. The PM's job is to understand what the user will respond to when they see a specific shade of green on a button. Strategy matters, but execution is molecular.
This belief—that consumer products live and die in the pixels—shapes how Duolingo builds. When Kansu joined in 2016, he was the first PM handed the monetization challenge. His claim to fame was putting ads on Duolingo, a move that earned both good and bad fame. But the attention to detail mattered more than the feature itself. Ad placement, load timing, and the color of the purchase flow all went through relentless A/B testing. The result: subscriptions now generate over 80% of Duolingo's revenue.
The design process starts with heavy usage of the app. Most ideas come from the team's own frustrations. When they felt Duolingo didn't teach speaking well enough, GPT-3 became the unlock. The prototype for Video Call with Lily took two to three months of iteration: first utility, then design, then perfection. Early builds had janky illustrations and error codes, but the core interaction—whether Lily could give smart answers—was validated before polish.
Consumer products live and die in the pixels. If a button that's the dark shade of green versus the light shade of green will make a difference in user behavior, a product manager should understand what that means.
— Cem Kansu
Figma helps teams decide if a direction is worth pursuing, but for experiential features, you have to experience them. Duolingo's product review process filters out mediocrity early. Teams show their work, get feedback, refine, and either prototype or ship an A/B experiment. There's subjectivity—one person finds Lily engaging, another doesn't—so Kansu brings high-judgment people together to align. If enough smart users say, "Yeah, this works better than anything else on Duolingo," the team moves forward.
Vibe-Coded Prototypes and the Death of Specs
Duolingo shipped Chess in eight months with no engineer. A product manager vibe-coded the entire experience. A designer created the visuals. They iterated on the prototype until Kansu and the review team agreed it felt like a Duolingo lesson. Only then did they staff a team with engineers. The timeline was half of what Math and Music required, with a third of the people.
This isn't an anomaly. It's the future Kansu is building toward. Traditional product development—idea to spec to design to engineering—is dead. AI coding tools eliminate the middle assembly line. Why write a spec when you can go from text to a working prototype? Smaller teams will move faster, and Duolingo will work on 10x more innovative bets.
Specs, like, you know, companies write product specs that 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. Why would you write specs if you can go from your idea to an exact prototype?
— Cem Kansu
Kansu tells his team to be AI-first, even when it feels artificial. He compares it to companies going mobile-first in 2012, when 1% of users were on mobile and 99% were on web. The bet felt stupid at the time, but it paid off in three years. Today's AI tools produce dinky web apps, but Kansu expects that gap to close in 12 months. Product managers who know how to harness AI to build will come out ahead.
The incentive structure reinforces this. Duolingo rewards people who automate repetitive tasks with recognition—public sharing and applause. Compensation follows. High performers who create more output with AI get promoted faster. The goal isn't just efficiency; it's cultural transformation. Roles will converge. Engineers will stretch into design and PM. Designers will vibe-code. PMs will do all of it. The traditional boundaries are blurring.
Growth Levers Compounded Over Eight Years
No single feature changed Duolingo's growth curve. It's been eight years of compounding improvements to a few core levers: streaks, leaderboards, push notifications, and better courses. The streak existed at launch, but it's evolved. It's now animated, shown prominently on the homepage and profile, and carries emotional weight. The higher the streak, the less users want to stop. It's a retention mechanic that compounds over time.
Leaderboards exploit human nature. When you see your name ranked against others, you don't want to be at the bottom. The weekly cycle resets the game, but the drive to collect XP and win keeps users doing more lessons. Push notifications are another lever, and Duolingo's are famously unhinged. Practice reminders are written in Duo's voice. If you ignore them for a few days, the app doesn't spam you forever—it sends one last passive-aggressive note: "Hi, it's Duo. These reminders don't seem to be working. We'll stop sending them for now."
We use the term unhinged as a very positive thing. When someone says, 'Oh, that change is unhinged,' at Duolingo, that means you've done a good job. Our app icon is one. When Duo does something really weird in our app, we're like, 'Yeah, cool, that's really unhinged.'
— Cem Kansu
The copy reads like a text from a pissed-off significant other. It's gone viral on Twitter, giving Duo even more character. Duolingo optimizes emoji usage, notification timing, and personalization. The team experiments constantly, but the principle remains: instill character and tone into every interaction.
Course quality matters too. Duolingo's French course isn't static. The team adds harder content, improves sections, and tailors difficulty to learners. The challenge is that not everyone starts from zero. Some users took two years of French in high school. Others learned from a cousin. Placing users at the right level based on what they know is hard, but it helps retention. First impressions matter—especially for brand-new beginners—but the experience has to scale for intermediate and advanced learners too.
The Goal Is Not 40 Subjects
Duolingo teaches four subjects: language, math, music, and chess. Kansu's goal isn't to reach 40 subjects. It's to make what they have 10x better. Options, he knows, can be an enemy. Focus matters in product. You can distract yourself by going too broad. If they find a great subject that's additive to Duolingo and believe they can do it really well, they'll add it. But breadth is not the strategy. Depth is.
This discipline shows up in Duolingo's monetization model. When Kansu joined in 2016, the company had no revenue. CEO Luis von Ahn had famously said Duolingo would never have ads, in-app purchases, or subscriptions. Kansu built all three, in that order. The company tested what worked: how much ads hurt retention, what kind of ads perform best, whether to show them at session start or end, whether to call the subscription Plus or Super, what color the purchase flow should be. There was no grand vision—just relentless experimentation.
The breakthrough was freemium. Duolingo charges for bells and whistles—no ads, top-tier support, more locations, higher-speed servers, multi-platform access—but never for the core content. Users who pay and users who don't learn the same amount of French. This unlocked subscriptions as 80% of revenue and kept the mission intact: access to education for everyone.
We stand for access to education, so we always wanted to provide the education component of Duolingo free and wanted to figure out how to charge for the bells and whistles.
— Cem Kansu
The model works because Duolingo doesn't over-optimize the free tier. The product is legitimately good. Conversion from free to paid is deliberately low, and Kansu is fine with that. The top of the funnel is the superpower. Don't mess with it. Give people a fantastic free experience, and money will come. The same philosophy applies to international expansion: make sure the product works, localize properly, and build unhinged TikTok content in Japanese for Japan, not just English for the US.
The PM Who Learns AI Tools Wins
If you're 23 and entering product management today, Kansu's advice is blunt: learn AI tools in and out, and know they'll look different in four weeks. Experiment heavily with how AI tools build product. The landscape will shift fast, but the muscle you develop—going from idea to prototype without an assembly line—will define the next decade of product work.
Duolingo is already there. The team is AI-first. The Chess example proves it. Smaller teams, faster timelines, more innovation. The traditional product team—5 to 15 people with engineers, designers, PMs, data scientists, and marketers—will shrink. You might not need as many people to create a prototype or test a design. The roles will converge. The tools will enable hybrid people who can design, code, and strategize.
Kansu doesn't know if Figma becomes obsolete or if it evolves into the tool that takes you from text to prototype. But he's certain the middle steps—specs, handoffs, assembly lines—are dead. The future is smaller, faster, weirder. And at Duolingo, the highest compliment you can give a product change is still the same: "That's really unhinged."