Cem Kansu on How Duolingo Ships 10x Faster by Killing the Product Spec

The Product Spec Is Dead

Cem Kansu doesn't mince words: the traditional product development process is obsolete. At Duolingo, where he leads product for 90 million monthly active users, the old playbook of idea to spec to design to code is disappearing fast.

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 idea to an exact prototype?

— Cem Kansu

The unlock? AI-powered prototyping. When Duolingo tested GPT-3 for teaching speaking skills, the team went straight from "LLMs could help us teach speaking" to a working prototype of Video Call with Lily in months, not quarters. No lengthy spec documents. No multi-stage handoffs. Just rapid iteration from concept to code.

This isn't theoretical. Duolingo just shipped an entire chess course in 8 months with a team of two: one PM who vibe-coded the prototype and one designer. No engineers until late stage. Compare that to their earlier launches of Duolingo Math and Music, which required triple the headcount and double the timeline.

Consumer Products Live and Die in the Pixels

Kansu rejects the "PM as mini-CEO" trope outright. His mantra: consumer products live and die in the pixels. That means product managers need to understand pixel-perfect design, not just high-level strategy.

If a button that's like the dark shade of green versus the light shade of green will make a difference in user behavior, a product manager or a product designer, for that record, like people working on the product, should understand what that means.

— Cem Kansu

Duolingo's design process starts with heavy internal usage. Most ideas come from the team using the app obsessively. When they felt Duolingo didn't teach speaking well enough, that user pain became the impetus for Video Call with Lily.

The team uses Figma for directional feedback but prioritizes experiencing the actual product. For Video Call, they prototyped for 2-3 months in stages: first validating the core interaction (can Lily give smart answers?), then making it visually engaging, then perfecting execution. Each prototype was janky and incomplete, but it answered one critical question before moving to the next.

Bottoms-Up Innovation at Speed

The chess course exemplifies Duolingo's bottoms-up culture. A PM and designer prototyped the entire experience without executive mandate or engineering support. They kept iterating on the vibe-coded prototype until leadership said: "This is a pretty good Duolingo lesson." Only then did they staff engineers and ship to users 8 months later.

Kansu's bet: smaller teams with AI tools will let Duolingo work on 10x more innovative ideas simultaneously. Not 10x more subjects (they teach just 4: language, math, music, chess), but 10x better execution on what they already do.

The Hybrid Prototyper Is Coming

Kansu predicts the lines between engineer, designer, and PM will blur dramatically. He calls this emerging role the "prototyper": someone who can use AI tools to design, code, and ship without traditional handoffs.

Duolingo has already declared itself "AI first" as a product team, mirroring the mobile-first shift of 2012. Back then, betting on mobile felt artificial when only 1% of users were on mobile devices. But it was the right call. Kansu sees AI tools the same way.

Anyone that's going into products should bet on the same thing, which is you should build AI first, even though when you use these AI tools and say, all right, create me a great app, you end up with like a dinky web app. In my opinion, that will get solved in, I don't know, maybe 12 months.

— Cem Kansu

His advice to early-career PMs: learn AI tools inside and out, even though they change every 4 weeks. Duolingo rewards team members who automate repetitive tasks with recognition, faster promotions, and compensation. The company bakes AI productivity into performance reviews.

Right now, Kansu sees tools like Cursor delivering 10-20% productivity gains through autocomplete, not the 10x promised by hype cycles. The real unlock comes when text-to-pixel-perfect-design and design-to-shippable-code work seamlessly. He thinks that's 12 months away, not 5 years.

Never Say Never on Monetization

When Kansu joined Duolingo in 2016 as the first monetization PM, founder Luis von Ahn had publicly promised: no ads, no in-app purchases, no subscriptions. Ever. Kansu's job was to implement all three.

The original monetization plan was crowdsourced translation. Users would translate articles for CNN and other publishers as part of their exercises. It didn't scale. Duolingo was pre-revenue and needed a path forward.

The breakthrough was freemium done differently. Every education app locked content behind paywalls because it seemed obvious: people came to learn, so charge for learning. Duolingo did the opposite. All learning content stayed free. They monetized bells and whistles: no ads, unlimited hearts, offline access.

Even investors were skeptical. The common wisdom was to prioritize growth and retention over monetization. But Kansu proved you could do both. Tastefully implemented ads didn't kill retention. Subscriptions became the lion's share of revenue while maintaining Duolingo's mission of free education access.

When Kansu showed Luis early ad prototypes, the feedback was blunt: "These ads look like shit." But they tested, iterated on design, and shipped. The retention fears never materialized. The lesson: things change, and your view on monetization changes with them.