Duolingo: 148 new language courses in a year with generative AI

• Duolingo Launches 148 New Language Courses — Duolingo (пресс-релиз), 2025-04-30
• Duolingo launches 148 courses created with AI after sharing plans to replace contractors with AI — TechCrunch, 2025-04-30
• Duolingo CEO walks back AI-first comments: 'I do not see AI as replacing what our employees do' — Fortune, 2025-05-24
• Duolingo CEO admits his controversial AI memo 'did not give enough context' — Fortune, 2025-08-18

Background

Duolingo is the world's largest language-learning app and a public company (NASDAQ: DUOL) whose valuation depends directly on its ability to grow into new audiences. A Duolingo course is not a 'deck of flashcards' but thousands of exercises, interactive stories, and audio materials structured along the European CEFR scale. Historically each course was built by a dedicated team of learning designers and writers: the company's first 100 courses took about 12 years to develop.

By the mid-2020s Duolingo found itself in a position familiar to every content business: the product scales to millions of users instantly, but content production does not. Each new language pair (say, 'Japanese for Hindi speakers') required months of manual work, and the combinatorics of languages made the task endless: even 40 languages in both directions is fifteen hundred potential pairs.

On April 28, 2025 CEO and co-founder Luis von Ahn sent employees a memo declaring Duolingo an 'AI-first' company: AI use would become part of performance reviews, new headcount would only go to teams that could not automate further, and contractors would gradually stop being engaged where AI could generate the content. 'Without AI, it would take us decades to scale our content to more learners,' von Ahn wrote.

Two days later, on April 30, 2025, the company showed what those words meant: the announcement of 148 new language courses at once — the largest content expansion in Duolingo's history, created in under a year.

Problem

Manual course production did not scale, and this was not an abstract problem but a precisely measurable access problem. The most in-demand languages — Japanese, Korean, Mandarin — were available from only a few interface languages, mostly English. A Hindi or Portuguese speaker could not learn Japanese from their native language: hundreds of millions of potential learners were blocked by the absence of a course, not the absence of desire.

The bottleneck was neither ideas nor methodology — Duolingo had long since standardized what a course should look like: the CEFR level progression, exercise types, Stories for reading comprehension, DuoRadio for listening. The bottleneck was production speed: thousands of content units per language pair, each previously written and reviewed by humans.

The classic economics of the process ran into combinatorial explosion. Creating 'a Spanish course' is not enough — you need Spanish for speakers of English, Portuguese, Hindi, Mandarin, and so on, and each version is an adaptation, not a translation: hints, grammar explanations, and typical errors depend on the learner's native language. The manual pipeline that took 12 years to produce the first 100 courses physically could not cover that matrix.

For a public company with rising investor expectations this meant a strategic ceiling: audience growth was limited not by product or marketing but by how fast courses for new language pairs could appear.

Solution

Duolingo rebuilt production around three components: generative AI, internal content-creation tooling, and an approach the company internally calls 'shared content'. The idea of shared content is that the team creates one high-quality base version of a course — a gold standard with vetted methodology — and then rapidly customizes it for dozens of interface languages. AI does the mass generation and adaptation, while human experts set the methodology and review the output.

This addresses precisely the combinatorial problem: instead of building each 'Japanese from Hindi' course from scratch, the company builds one reference Japanese course and replicates it across 28 interface languages. Previously such adaptation took nearly as much manual work as creating the course itself — LLMs made it fast and cheap while keeping quality control with the experts.

Where the company started matters. The new courses cover CEFR A1–A2 — the beginner rungs, where content is most templated and the structure repeats from language to language: basic vocabulary, simple grammar, typical dialogues. Advanced content, where a methodological error costs more, was announced separately for 'the coming months'. And the new courses shipped with the signature formats included — Stories for reading and DuoRadio for listening — so AI production did not mean a stripped-down product.

In parallel, von Ahn described the organizational side: 'AI-first' in his memo meant structural changes in how teams work — from hiring to performance reviews. The memo said outright that the company would 'gradually stop using contractors to do work that AI can handle', and that new headcount would only be granted if a team could not automate further. Later, amid the public reaction, the company clarified: no full-time employees were laid off, hiring continues at the same pace, and teams got workshops, advisory councils, and experimentation time to adapt to AI.

Duolingo does not publicly disclose which models power the generation — only that it is generative AI combined with proprietary tooling and human review. The company deliberately describes the production system, not the stack: its competitive advantage is methodology and learning data, not a particular LLM.

Result

On April 30, 2025 Duolingo announced 148 new language courses at once — the largest content expansion in company history, created in under a year. The course catalog more than doubled. The seven most popular non-English languages — Spanish, French, German, Italian, Japanese, Korean, and Mandarin — became available from all 28 interface languages, opening learning to over a billion potential learners. Von Ahn framed the speed comparison himself: 'Developing our first 100 courses took about 12 years, and now, in about a year, we're able to create and launch nearly 150 new courses. This is a great example of how generative AI can directly benefit our learners.'

The launch had a second, less celebratory side — a communications crisis. The 'AI-first' memo about phasing out contractors, published two days before the course announcement, triggered a wave of criticism: users declared they were deleting the app, and the comments under the company's TikTok and Instagram posts turned into a stream of anti-AI sentiment. In May 2025 von Ahn issued a clarification on LinkedIn: 'To be clear: I do not see AI as replacing what our employees do (we are in fact continuing to hire at the same speed as before). I see it as a tool to accelerate what we do.' In August 2025 he admitted in an interview that the memo 'did not give enough context' and stressed that no full-time employees had been laid off.

Frame the result correctly: 148 courses and a doubled catalog are verifiable product facts; but the company published no quality metrics for the new courses (learner retention, level progression) at announcement time, and part of the user criticism targeted exactly the quality of AI content.

In our view, the Duolingo case is the clearest demonstration of where generative AI gives a content business maximum leverage: not 'writing instead of people' but replicating one vetted gold standard across dozens of localizations. The '100 courses in 12 years versus 148 in a year' formula became a public speed benchmark for the whole EdTech industry — while honestly comparing tasks of different complexity: the first 100 courses included building the methodology from scratch, whereas the 148 new ones replicate it at beginner levels.

The second lesson is the price of words. The difference between 'AI accelerates our teams' and 'we will replace contractors with AI' cost the company weeks of public crisis over essentially identical changes. For executives planning an AI transformation, communications strategy is not an appendix to the project but part of it — and the Duolingo case should be studied alongside Klarna's reversal: the market punishes radical rhetoric even when the operational results are real.

Lessons learned

1. Generative AI's real lever in a content business is not 'writing instead of people' but replicating one gold standard across dozens of localizations (shared content): it solves the combinatorial explosion of language pairs.
2. Comparing speeds (100 courses in 12 years vs ~150 in a year) is the most convincing public metric format: the order of magnitude is visible without internal data.
3. AI scaling started at CEFR A1–A2, where content is most templated — not at advanced levels where methodological mistakes cost more; a transferable principle for choosing the entry point.
4. Humans stay the owners of methodology and quality control: AI generates and adapts, experts set the gold standard and review — otherwise you replicate errors, not quality.
5. Plan the communications risk of an 'AI-first' strategy as part of the project: the contractor memo overshadowed the product launch itself and required two public CEO clarifications.
6. Wording decides: 'AI is a tool to accelerate our teams' and 'we will replace contractors with AI' describe the same change, but the second version cost the company a wave of app deletions.
7. Investing in internal content-production tooling pays back much faster than hiring per new course — but publish quality metrics alongside speed, or skeptics will fill the vacuum for you.

Frequently asked questions