I Didn’t Think It Was Possible, But They’re Making Duolingo Worse

You’ll remember my rant against Duolingo, which has now turned one of Aftermath’s Discord channels into a regular vent session as we share the language learning app’s latest follies. (My recent complaint: I pop in to do one lesson and it keeps foisting XP boosts on me, a naked attempt to keep me on the app that absolutely works, because I am a rube.) Well, I’m back with a new complaint, which is the harder pivot to AI Duolingo CEO Luis von Ahn announced yesterday.

In a message to staff also posted to the company’s LinkedIn, von Ahn writes that “Duolingo will remain a company that cares deeply about its employees,” then says that the company will be replacing more contractors with AI, writing, "We’ll gradually stop using contractors to do work that AI can handle." This is part of the company’s decision to become “AI-first,” a nonsense buzzword that even von Ahn can’t pitch without admitting it sucks, writing (emphasis mine) “Being AI-first means we will need to rethink much of how we work… [W]e can’t wait until the technology is 100% perfect. We’d rather move with urgency and take occasional small hits on quality than move slowly and miss the moment.”

This strategy will also involve considering AI in hiring and performance reviews, whatever that means, and that “headcount will only be given if a team cannot automate more of their work.” While von Ahn writes that "This isn’t about replacing Duos with AI" (contractors are indeed not staff) he does write that "It’s about removing bottlenecks so we can do more with the outstanding Duos we already have," which reads to me like a way of fancying up the "do more with less" afflicting countless industries these days.

In his note, von Ahn seems to claim that these decisions are in service of Duolingo’s mission:

To teach well, we need to create a massive amount of content, and doing that manually doesn’t scale. One of the best decisions we made recently was replacing a slow, manual content creation process with one powered by AI. Without AI, it would take us decades to scale our content to more learners. We owe it to our learners to get them this content ASAP.

Rather than use the revenue from Duolingo’s nearly 10 million paid subscribers to hire more experts to create–I cannot stress this enough–learning content that requires subject matter expertise, von Ahn seems to believe that quantity is the top goal, and that its users deserve a flood of what even he admits could be shit. This certainly squares with my own experience of Duolingo, which employs every cheap gamification trick in the book to simply keep me in the app and push me to pay.  

On LinkedIn, Duolingo’s chief engineering officer Natalie Glance shared a Slack message she wrote to the engineering team explaining how AI will fit into their work. This message features new directives such as “AI will be everywhere in our product” and “productivity expectations will rise.” Here’s my personal favorite:

Start with AI for every task. No matter how small, try using an AI tool first. It won’t always be faster or better at first–but that’s how you build skill. Don’t give up if the first result is wrong.

Again, these people are literally admitting AI won’t work, but are still asking engineers now laboring under heightened productivity expectations to spend their even-more-precious time managing this sub-par tech. "That's how you build skill" puts the AI tools' failing back on the user, the "have fun staying poor" of the AI world, and asks engineers to troubleshoot AI when they could instead be using their real live human skill to do the jobs they were hired for.

That Duolingo execs can write all this and still insist the company prioritizes humans, whether that’s workers or users, is a staggering level of AI brainrot. None of this helps people. Workers, if they still have jobs, will be asked to pivot to clumsy tools. Users will be fed content Duolingo itself admits could have mistakes; in the language I’m learning, Irish, this is already a problem, with the computerized voices mangling the pronunciation to such an extent that I live in dread of the app giving me listening exercises because I often have no clue what it’s attempting to say. All of this means what AI always means: creating more–God help me–content with less people, assuming workers and users are so stupid that we’ll lap up whatever shit is shoved in our faces or crammed into our workflows because the people at the top can’t imagine we’re not as enthralled with the latest fad as they are. 

I’m new enough to Duolingo that giving up my streak or uninstalling the app isn’t a big ask, though I’ll also admit that I’m loathe to give up on all the new vocabulary it’s teaching me. Luckily, I have options: my public library gives me free access to Mango Languages, and I know other libraries have Rosetta Stone or other programs. I’ve recently been experimenting with user-made Anki decks; the program is much clunkier than Duolingo, but way more customizable and robust. For those of you learning less obscure languages than Irish, I’m sure there’s an even greater wealth of resources out there.

Update, 5/2/25, 3:15pm-- According to journalist Brian Merchant, Duolingo has "replaced up to 100 of its workers—primarily the writers and translators who create the quirky quizzes and learning materials that have helped stake out the company’s identity—with AI systems." A Duolingo contractor who spoke to Merchant anonymously told him that their job focus became training the AI system rather than writing content, and that "We had been working with their AI tool for a while, and it was absolutely not at the point of being capable of writing lessons without humans.”

The worker told Merchant,

First, the AI output is very boring. And Duolingo was always known for being fun and quirky. Second, it absolutely makes mistakes. Even on things that you would think it could get right. The AI tools that are available for people who pay for Duolingo Max often get things wrong—they have an ‘explain my mistake’ tool that often will suggest something that’s incorrect, sometimes the robot voices are programmed to speak the wrong language.