Duolingo scraps AI performance reviews after employees push back on flawed metrics

Duolingo Abandons AI Performance Reviews

Duolingo has officially ended its controversial practice of evaluating employees based on AI usage, marking a significant AI U-turn for the language-learning platform 1

1

. CEO Luis von Ahn acknowledged the misstep during a recent Silicon Valley Girl podcast episode, admitting that the company's AI-first strategy created more confusion than clarity among staff 2

2

.

The reversal comes just months after von Ahn announced plans in April 2025 to transform Duolingo into an "AI-first" company, which included integrating AI into operations and eventually eliminating contract workers. A key component involved factoring AI tool adoption into employee performance reviews, a practice already implemented by tech industry giants including Meta, Google, Microsoft, and Shopify 1

1

4

4

.

Employee Pushback Exposes Flawed Metrics

The policy quickly sparked internal resistance as employees questioned what exactly was being measured. Workers asked whether the company wanted them to use AI for AI's sake, or whether using AI was more important than delivering quality work outcomes 1

1

. "It felt like rather than being held accountable for the actual outcome, we're trying to just push something that in some cases did not fit," von Ahn explained 2

2

.

The CEO's admission reflects a fundamental tension in how companies approach AI at work: counting usage is easy, but measuring actual impact on employee productivity and job quality proves far more complex 2

2

. Von Ahn now emphasizes that "the most important thing in your performance is that you are doing whatever your job is as well as possible. A lot of times AI can help you with that. But if it can't, I'm not going to force you to do that" 1

1

.

A Pattern of AI Overcorrection

This isn't Duolingo's first retreat from aggressive AI adoption. In January 2024, the company laid off 10% of contract workers due to automation 1

1

. When von Ahn announced the AI-first strategy in April 2025, urging teams against hiring more human workers unless automation proved impossible, users responded with social media backlash threatening to delete the app 1

1

3

3

. Von Ahn admitted he "did not expect the blowback" and backtracked within a week, stating that he doesn't view AI as a replacement for Duolingo employees 3

3

.

Despite the setbacks, von Ahn maintains that AI tools have enabled Duolingo to introduce 148 new language courses and that "a single employee is just way more productive now than they used to be" 3

3

. However, he concedes AI tools struggle with generating large numbers of narratives and that "it's not yet the case" that AI can code better than humans, with debugging mistakes often taking longer than traditional methods .

A Cautionary Tale for Tech Companies

Duolingo's experience serves as a cautionary tale as organizations rush to demonstrate AI adoption. Other companies are discovering similar limitations when integrating AI into operations. Klarna, once the most vocal AI advocate, started hiring humans again after finding AI chatbots offered "lower quality" output 1

1

. Take-Two recently conducted a layoff of its AI team, including its head of artificial intelligence, just two months after publicly embracing generative AI 1

1

.

The shift in performance evaluations signals a broader reckoning across the tech industry about sustainable AI implementation. While AI remains part of Duolingo's strategy, it's no longer a proxy for measuring success 2

2

. Companies tracking AI usage with an eye on productivity gains must now confront whether forcing automation into every workflow risks prioritizing tool adoption over actual results.