No Priors Ep. 114 | With Duolingo CEO Luis von Ahn

Moving from 30-minute to 2-minute lessons dramatically increased engagement, not by reducing total time spent, but by lowering the psychological barrier to starting; people will often chain many short sessions once they’ve begun.

Von Ahn argues that 90% of learning outcomes depend on actually showing up and putting in hours, so Duolingo intentionally borrows engagement mechanics from games and social apps to keep mostly unmotivated users coming back.

Simple streak counters and a passive-aggressive “we’re going to stop sending reminders” notification proved unexpectedly effective at reactivating users, revealing how loss aversion and perceived abandonment can be harnessed to sustain habits.

Large language models now generate most of Duolingo’s course content and enable realistic, judgment-free conversation practice, allowing the company to offer many more language pairings and richer experiences than human authoring alone.

Duolingo’s models predict performance per exercise and aim to give users items they’ll get right roughly 83% of the time—high enough to feel competent but not so high as to be boring—maximizing enjoyment and persistence over hundreds of hours.

Leaning into the internet’s dark memes about the Duolingo owl and creating absurd TikToks built outsized awareness and affection; being an education company gives them more leeway to be “unhinged” without heavy backlash.

Von Ahn expects classrooms to increasingly rely on AI for individualized instruction while teachers handle care and supervision, but bureaucracy, regulation, and cultural norms mean the shift will unfold over many years rather than overnight.