‘I did not expect the blowback’

Is Luis von Ahn’s job as chief executive of Duolingo vulnerable to displacement by artificial intelligence?

“I’m not going to claim CEOs are that special,” he says. “It’s just somebody has to tell others . . . ‘This is where we’re going.’ And AI is not particularly good at that yet.”

Judging by recent controversy around the gamified language-learning app he co-founded, it is not just AI that can misjudge communication. Last month, von Ahn shared on LinkedIn an email he had sent to all staff announcing Duolingo was going “AI-first”. “I did not expect the amount of blowback,” he admits.

“Unbelievable,” wrote one LinkedIn commenter, “Cancelling my account NOW.” “Well, there goes my 1,098 day streak,” posted another.

The chief executive says many social media users mischaracterised the changes as though “Duolingo has no employees, we have fired everyone and everything is being controlled by a massive AI”.

He attributes this anger to a general “anxiety” about technology replacing jobs. “I should have been more clear to the external world,” he reflects on a video call from his office in Pittsburgh. “Every tech company is doing similar things [but] we were open about it.”

He insists the reaction from staff and contractors was far more measured. “Nobody misunderstood”, he says, because when Duolingo started 14 years ago, “we knew this was going to be a technology-forward company”. Staff raised most questions over the proposal that their performance review was to include an evaluation of their use of AI. The details have yet to be worked out. 

Since the furore, von Ahn has reassured customers that AI is not going to replace the company’s workforce. There will be a “very small number of hourly contractors who are doing repetitive tasks that we no longer need”, he says. “Many of these people are probably going to be offered contractor jobs for other stuff.”

Duolingo is still recruiting if it is satisfied the role cannot be automated. Graduates who make up half the people it hires every year “come with a different mindset” because they are using AI at university.

The thrust of the AI-first strategy, the 46-year-old says, is overhauling work processes. He personally is already outsourcing some tasks to AI, including Excel calculations. He wants staff to explore whether their tasks “can be entirely done by AI or with the help of AI. It’s just a mind shift that people first try AI. It may be that AI doesn’t actually solve the problem you’re trying to solve . . . that’s fine.”

The aim is to automate repetitive tasks to free up time for more creative or strategic work.

Examples where it is making a difference include technology and illustration. Engineers will spend less time writing code. “Some of it they’ll need to but we want it to be mediated by AI,” von Ahn says. In exchange for productivity gains, they will be expected to devote 10 per cent of their time to learning. Similarly, designers will have more of a supervisory role, with AI helping to create artwork that fits Duolingo’s “very specific style”. “You no longer do the details and are more of a creative director. For the vast majority of jobs, this is what’s going to happen.”

He insists AI is primarily about speed. “Part of the reason we only [teach] 40 languages is because adding another language is a lot of work. Pretty soon we’ll be able to multiply the number of languages we teach by a lot.” The technology is changing “very fast . . . A lot of things are becoming possible on a daily basis.” That means putting up with imperfections in the short term, with the expectation they will be ironed out. In Duolingo’s early days it used automated speech instead of actors to sound out words. “It sounded a little robotic,” he explains. But over time, the technology improved.

Last year, Duolingo introduced a new feature to its premium (Max) subscribers, enabling users to converse with an avatar called Lily, a fictional moody teenager with a “sassy personality”, “signature sarcasm” and eye rolls. Over time, the chatbot learns more about the user’s preferences and abilities, so the relationship becomes more personalised. Von Ahn is open-minded about the future of such para-social interactions, mentioning a recent trip to China, where he saw that “a lot of people are talking to their AI friend . . . and it didn’t seem that bad, honestly.

“You can always think of these Black Mirror [the dystopian science fiction series] episode-type situations, right? Some people are going to spend a significant amount of time socially talking to AI, that’s just inevitable.”

Other societal implications for AI, such as the ethics of stealing creators’ copyright, are “a real concern”. “A lot of times you don’t even know how [the large language model] was trained. We should be careful.” When it comes to artwork, he says Duolingo is “ensuring that the entirety of the model is trained just with our own illustrations”. He is less worried about translation because sentences are usually common language, such as “the boy runs”, rather than excerpts from authored books.

Duolingo has grown rapidly since it listed in the US four years ago. Its first-quarter results showed it had 10.3mn paying subscribers, a 40 per cent increase on the year before, helping lift revenues 38 per cent to $230mn, and net income up 30 per cent to $35.1mn. Users fall into two camps: those learning a language to get a job or move to another country, and those doing it as a hobby. The enjoyment factor means he sees his main competitors as TikTok and Instagram, rather than other language apps. Duolingo has expanded to offer chess, music and maths.

Raised in Guatemala City by his single mother, a paediatrician, von Ahn speaks English and Spanish “perfectly well”. His French is good enough to understand TV shows without subtitles “but my pronunciation sucks”. He is “pretty good at Portuguese” and a beginner in Swedish “because my wife is Swedish” and in Japanese.

Von Ahn moved to the US to study maths, followed by a PhD in computer science at Carnegie Mellon University, and research in cryptography. He started a business, ReCaptcha, which screens for bots and spam, and sold it in 2009 to Google for a figure enabling retirement before realising, as he later told the Financial Times, he “would get bored”.

Perhaps his personal wealth contributed to his lack of urgency to make money at Duolingo? It took him about six years to tell staff the company’s losses were unsustainable. “You would be surprised how difficult that was. It took about six months to convince this early group of employees that a, making money is not evil, and b, we need to do that if we’re a company. I took the whole social warrior/communist thing too far, and the reality is had we started monetising three years early, we would be much ahead.”

The shift to going public has “been great”, he says. The rigour required has improved processes such as financial forecasting. His mother, who now lives in one of his properties in Pittsburgh, “doesn’t really understand what my company does [but] she somehow knows the stock price.”

Aside from its soft toy owl merchandise, which is made in China, Duolingo is unaffected by US tariffs. “We have two options: either increase the price or make the owl smaller. It’ll probably be smaller.” While the Trump administration is on a campaign to roll back employer initiatives on race and gender, von Ahn says a diverse workforce is essential because “we have users in every single country in the world.” He says its international employees are “nervous because who knows what’s going to happen?”  

After we speak, von Ahn returns to LinkedIn to apologise for his previous lack of clarity. “I don’t know exactly what’s going to happen with AI,” he writes. “But I do know it’s going to fundamentally change the way we work . . . AI is creating uncertainty for all of us, and we can respond to this with fear or curiosity.”