Duolingo Admits Mistake After a Year: No More Evaluating Employees on AI Usage
Duolingo has reversed its evaluation of the use of artificial intelligence in employees' performance reviews. Less than a year after the announcement of the "AI-first" strategy, CEO Luis von Ahn publicly admitted that measuring how much employees use AI was a misstep, and thus decided to abandon the metric.
The 2025 Memo and the Backlash
On April 28, 2025, von Ahn distributed a memo to employees declaring Duolingo an "AI-first" company, with three concrete operational implications: hiring would prioritize candidates with AI skills; contractors would gradually be replaced by AI for automatable tasks; and the use of artificial intelligence tools would become a formal parameter in performance reviews. The communication was later shared on the company's LinkedIn profile, amplifying the media coverage well beyond internal boundaries.
The reaction was immediate and more intense than expected. Some app users threatened to delete the app, worried that increased automation would degrade the learning experience. Von Ahn himself admitted a few months later that he hadn’t anticipated such a significant backlash. In August 2025, he issued a further clarification, stating that the memo had been misunderstood and that there were no plans to lay off full-time employees.
The Reversal on Performance Reviews
On April 10, 2026, as a guest on the podcast "Silicon Valley Girl," von Ahn precisely described the dynamics that led to the turnaround. After the company started using AI as a evaluation criterion, employees began to ask, "Do you want us to use AI just for the sake of using it?" It was a sign that something was wrong with the framework. "In the end, we backtracked," von Ahn said. "The most important thing in your performance is that you do your job in the best way possible. Often AI can help you with this. But if it can't, I’m not going to force you to use it."
The CEO acknowledged that the original metric risked rewarding activity rather than results. "It seemed that instead of being held accountable for actual outcomes, we were trying to push something that in some cases didn’t fit," he stated. From now on, evaluations will return to measuring the quality of work and concrete outcomes, regardless of the tools used to achieve them.
Internal Technical Doubts
Von Ahn also shared internal data that explain at least in part the revision of approach. AI has proven less reliable than expected in two key areas for Duolingo: story writing for courses and debugging automatically generated code. "The code written by AI can be difficult to fix and isn't consistently reliable for story writing," he said on the podcast. This position sharply diverges from the prevailing rhetoric in the tech sector, where AI-assisted coding is often presented as already mature for production.
This assessment aligns with findings from a global survey conducted by WalkMe, a subsidiary of SAP: more than a third of employees surveyed stated they deliberately skip using AI on certain tasks because it slows down workflow or takes more time than traditional approaches. This data suggests that top-down pressure for adoption might generate a form of silent resistance rather than a real transformation of work processes.
An Exception in the Tech Landscape
Duolingo's position stands out sharply compared to other big tech companies. Meta has introduced "AI-driven impact" as a formal criterion in performance reviews starting in 2026, with a system that evaluates employees based on how effectively they use AI to generate results and build productivity-enhancing tools. Until recently, Meta also had an internal leaderboard of the top 250 AI users based on tokens consumed at a corporate level, initiated by the employees themselves.
Google, for its part, has communicated to non-technical employees that AI usage in workflows is expected, and in some cases, it may affect evaluations. On the positive incentives side, the marketing automation platform Omnisend announced in April 2026 that employees classified as "outstanding AI users" would receive a raise between 2% and 4%. However, the criteria are results-oriented: savings in time and money generated from using AI, tangible results obtained via automated workflows, and adoption of those workflows by the rest of the team. An approach that, unlike simply measuring usage, aims to quantify, if it exists, the value produced.
In fact, Duolingo has acted ahead of the realization that many companies will likely need to make: enforcing AI adoption metrics without tying them to measurable outcomes produces operational confusion, not transformation. The company’s spokesperson reiterated the official position: "The work of our teams depends on human judgment, expertise, and creativity. AI tools assist that work; they do not make decisions and do not replace the people who build Duolingo."