Duolingo drops AI usage metric in employee performance reviews

Summary

This article addresses the reversal of Duolingo's policy of incorporating AI usage as a formal metric in employee performance reviews. The author reports that Duolingo CEO Luis von Ahn announced the original 'AI-first' policy on April 28, 2025, which tied employee performance assessments to how effectively they used AI tools. The article's central argument is that the policy generated significant internal and external backlash, prompting a formal reversal. Key evidence includes direct quotes from von Ahn on the Silicon Valley Girl podcast acknowledging that employees felt evaluated on AI adoption for its own sake rather than on outcomes. The article contextualises this reversal against a broader industry trend, citing Meta's internal AI usage leaderboard and Omnisend's AI-linked salary increases as counterexamples, while also referencing a WalkMe/SAP global survey finding that over one-third of employees actively avoid AI tools. The implied conclusion is that mandating AI adoption metrics in performance systems may produce resistance and misaligned incentives when decoupled from meaningful performance outcomes.

Strengths and limitations

Strengths: The article is grounded in direct, attributable quotes from the Duolingo CEO across two distinct public forums, providing credible primary sourcing. It also balances the Duolingo reversal against contrasting corporate practices at Meta and Omnisend, avoiding one-sided framing. Limitations: The WalkMe/SAP survey is cited secondhand via a Fortune report, with no sample size, methodology, or geographic breakdown provided, limiting its evidentiary weight. The article does not explore whether Duolingo's revised stance has been formalised in updated HR policy documentation or remains a CEO-level verbal commitment. The Omnisend case is described without corroborating employee or outcome data. Bias: The article leans toward framing the AI-usage metric as a failed initiative, with the reversal presented as a natural corrective — alternative perspectives on the original policy's merits are absent.

What this implies

The Duolingo case signals a potential limits-testing moment for AI-linked performance metrics: when AI adoption is measured independently of job outcomes, it risks creating compliance-driven behaviour rather than genuine productivity gains. The divergence between Duolingo's reversal and the approaches taken by Meta and Omnisend points to an unsettled industry consensus on how AI adoption is appropriately incentivised within PMS design. Employee resistance data from the WalkMe survey suggests that AI integration in performance frameworks may require differentiated treatment by role type rather than uniform adoption mandates.

Key points

• Duolingo formally reversed its policy of assessing employees on AI usage metrics after internal feedback indicated the approach incentivised AI adoption for its own sake rather than for meaningful job performance outcomes.
• CEO Luis von Ahn acknowledged AI's current limitations, noting that AI-generated code can be difficult to debug and that AI is not yet consistently superior to humans in areas such as coding and content creation.
• A WalkMe/SAP global survey found that more than one-third of employees avoided AI tools for certain tasks, citing workflow disruptions and increased time requirements, suggesting that mandated AI adoption may conflict with actual productivity.

What to take away

• Duolingo's experience illustrates that embedding AI usage as a standalone performance metric — separate from outcome-based measures — can generate employee resistance and public backlash.
• Organisations linking AI adoption to compensation or performance ratings, as seen at Omnisend and Meta, are operating in contrast to Duolingo's revised approach, which prioritises job outcomes over tool usage compliance.