What ‘AI first’ looks like at Duolingo

On Wednesdays, we wear pink and on Fridays, we play with AI…if you work at the language-learning company Duolingo, at least.

This past summer, Duolingo employees, known internally as Duos, have spent their Fridays learning how to use AI tools to improve their workflows. Natalie Glance, chief engineering officer at the company, told IT Brew that Duolingo has set aside two-hour time blocks for its workforce every week and developed structured learning materials to better improve employees’ understanding of these tools.

“In addition to the Fridays, we also have a weekly or biweekly hour-long engineering AI assembly where people present updates about new tools, or give demos of successes they’ve had, or things that work [and] things that haven’t worked,” Glance said.

All in on AI. The Friday sessions are not Duolingo’s first AI rodeo. In April, the language-learning company posted an all-hands email penned by CEO Luis von Ahn to LinkedIn that officially announced it would be going “AI first.”

At the time, von Ahn said the company would slowly phase out contractors used for work that AI could perform, only increase headcount if a team was unable to automate more of its work, and consider using AI during performance reviews and the hiring process. Duolingo’s recent rollout of dedicated time for employees to learn AI—known internally as “FrAIdays”—is a way of working toward its “AI-first” status.

“Being AI first means we will need to rethink much of how we work. Making minor tweaks to systems designed for humans won’t get us there. In many cases, we’ll need to start from scratch,” said von Ahn, who then publicly backtracked on the company’s AI vision in response to backlash it received, clarifying, “I do not see AI as replacing what our employees do.”

In a recent New York Times interview on the topic, von Ahn explained, “What will probably happen is that one person will be able to accomplish more, rather than having fewer people [employed].”

AI (and security) first! Glance said the adoption of AI tools at Duolingo has “taken off.” On an average day, 80% of Duolingo’s engineers are using an AI tool, she said.

As more employees continue to experiment with AI tools, Duolingo has taken a measured approach as to how it approves certain tools for internal usage. For instance, Glance said AI tools used to create production-level code must go through a security review at Duolingo, while AI tools used for prototyping new features go through a “lightweight” approval process, with a limited set of use cases when approved.

For AI coding tools, Glance noted the company has been “willing to have several different products” it works with.

“For example, Claude Code just burst onto the scene and everybody wanted to be able to use it, so we really fast-tracked the security review for that and the entire approval process and getting the contract signed,” Glance explained. While Duolingo has traditionally secured one-year contracts for AI tools, Glance said she has been in favor of six-month contracts because of how rapidly innovation is occurring in the market.

Experimentation and the payoff. Harley Sugarman, founder and CEO of the security platform Anagram, told IT Brew that experimentation is key for companies looking to go all in on AI. While some companies may choose to take a “strict” approach to the exploration phase, he believes it is more beneficial to give employees freedom over the tools they can trial and use in their workflows.

“Letting developers experiment with that is really important because if you don’t, you are just going to end up having them sort of do it in the shadows,” Sugarman said.

Since allowing employees to experiment with AI tools, Glance said she has seen interesting use cases throughout Duolingo. She recalled one intern, for example, who was able to use Cursor to finish a project five times faster than if she had completed it on her own.

“That’s pretty remarkable for an intern.”