Duolingo Goes AI-First in Push for Scalable Learning

• By John K. Waters
• 05/06/25

Back in 2012, Duolingo co-founder Luis von Ahn defied the prevailing wisdom of building desktop-first experiences and pushed his nascent language learning startup onto mobile. A year later, Duolingo was Apple's iPhone App of the Year. Fast forward to 2025, and von Ahn is making another bet. This time, it's on artificial intelligence — and it's potentially even more transformative.

"We're making a similar call now, and this time the platform shift is AI," von Ahn wrote in a company-wide memo posted publicly on LinkedIn. "AI is already changing how work gets done. The worst thing you can do is wait."

Duolingo, the Pittsburgh-based company behind the chirpy green owl and more than 500 million users, has officially declared itself "AI-first." But beneath that buzzy label lies a seismic restructuring — one that shifts roles, reshapes how hiring and performance are evaluated, and redefines the relationship between people and technology at Duolingo.

From Mobile-First to AI-First: A Cultural Reset

"AI-first" at Duolingo doesn't just mean sprinkling generative tech over existing workflows. It means dismantling them. The memo outlines sweeping changes: phasing out contractors in favor of AI tools, evaluating employee performance based on AI adoption, and restricting headcount growth unless teams can prove that automation isn't an option.

In the hands of a less charismatic founder, this might read like a Silicon Valley bloodletting. But von Ahn frames it as a kind of creative liberation.

"This isn't about replacing Duos with AI," he insisted. "It's about removing bottlenecks so we can do more with the outstanding Duos we already have."

That framing doesn't obscure the stakes. Duolingo will "gradually stop using contractors to do work that AI can handle," echoing a similar move last year when the company cut about 10% of its contract workforce after integrating AI into its translation pipeline. Now, AI will be enlisted to help with everything from content creation to hiring decisions to performance evaluations.

Scaling Learning, One AI at a Time

The rationale? Scaling. The same way mobile helped Duolingo go viral, von Ahn believes AI can democratize quality education at a planetary scale. "To teach well, we need to create a massive amount of content," he wrote. "Doing that manually doesn't scale."

AI, he argued, is the only realistic path forward. It's already being used to power features like Video Call, where users interact with AI-powered tutors like the deadpan, goth-inspired Lily. And Duolingo's content production — once a slow, editorial slog — is now increasingly generated by machine learning pipelines.

This isn't just about speeding things up. Von Ahn sees AI as a path toward pedagogical parity with human tutors — replicable, scalable, and always available.

Constraints as a Catalyst

To get there, von Ahn said, Duolingo must embrace "constructive constraints." Performance reviews will now include metrics for AI utilization. Hiring? Only if automation fails. Functions across the company are being asked to rebuild from the ground up.

"Making minor tweaks to systems designed for humans won't get us there," the memo reads. "In many cases, we'll need to start from scratch."

It's an echo of Shopify CEO Tobias Lütke's recent directive that employees justify new hires by proving AI couldn't do the job. It's also a signal that a new corporate orthodoxy is emerging — one where AI fluency is table stakes for survival.

The Uneasy Trade-Off

Of course, that new orthodoxy has its critics. On social media, some users lambasted Duolingo's shift as a betrayal of its quirky, accessible brand. Others, perhaps more resigned, suggested this was inevitable.

"When a CEO says they want more AI, it means they want more money in their bank account," one X user quipped.

There's also the question of efficacy. In a working paper released earlier this month ("Large Language Models, Small Labor Market Effects,") economists Anders Humlum and Emilie Vestergaard looked at the labor market impact of AI chatbots on 11 occupations, covering 25,000 workers and 7,000 workplaces in Denmark in 2023 and 2024. The researchers found that AI tools reduced actual work hours by just 2.8% — far less than the industry hype suggests. Critics warn that Duolingo's urgency could come at the cost of quality, nuance, and human insight.

Von Ahn seems unbothered. "We'd rather move with urgency and take occasional small hits on quality than move slowly and miss the moment," he wrote.