Duolingo just admitted what most companies won't: forcing employees to use AI because you can is management theater, not strategy.
The Summary
- Duolingo CEO Luis von Ahn reversed the company's policy of evaluating AI use in performance reviews, a year after announcing it as part of an "AI-first" mandate
- Employees pushed back with the right question: "Do you just want us to use AI for AI's sake?"
- The shift signals something bigger: outcome-based work beats metric-based theatrics, even at companies betting everything on AI
The Signal
In April 2025, von Ahn sent a memo positioning Duolingo as an "AI-first" company, complete with "constructive constraints" that would track employee AI use in performance reviews. A year later, he walked it back on the Silicon Valley Girl podcast. The reason? It stopped being about results and started being about checking boxes.
"It felt like, rather than being held accountable for the actual outcome, we were trying to just push something that in some cases did not fit," von Ahn said. Translation: we created a system where people gamed the metric instead of doing their jobs better.
"Do you just want us to use AI for AI's sake?"
This is the collision point between executive AI mandates and actual work. Here's what happened at Duolingo:
- Executives announce AI-first strategy with quantifiable tracking
- Employees start using AI not where it helps, but where it counts toward reviews
- Work quality becomes secondary to AI usage stats
- Leadership realizes they've built a perverse incentive system
This matters because Duolingo isn't some laggard company. They've been aggressive on AI, building language learning tools that actually use the technology meaningfully. Von Ahn faced real backlash for the original memo. Walking it back now isn't about being anti-AI. It's about learning what every company rushing to add "AI usage" KPIs needs to understand: tools serve outcomes, not the other way around.
The broader pattern: white-collar work is splitting into two camps. One camp uses AI where it multiplies their output. The other uses AI to hit metrics that make executives feel like they're "doing AI." Duolingo just figured out which camp creates actual value.
The Implication
Watch for more companies to quietly back away from AI-usage metrics in performance reviews over the next year. The ones that don't will select for employees who are good at appearing to use AI, not employees who are good at their jobs. That's an expensive filter to run.
If you're managing people, the lesson is clear: measure the work, not the tools. If AI makes someone 10x faster at their job, great. If it doesn't, forcing it creates busy work that looks like progress. Von Ahn just saved you from learning this the hard way.