Duolingo will replace contract workers with AI
theverge.com
theverge.com
Archived here: https://archive.is/zqk5z
If I was an engineer at a company that made this announcement I would not be feeling great right now. The claims that writing code will become a smaller part of our jobs and that productivity expectations will rise set off some alarm bells.
Some of the statements like "For example, we know that large language models work best with context" are alarming, as if the people writing this announcement have a very elementary understanding of how LLMs work but are making drastic policy changes based on their limited understanding.
Imposing rules on developers like the requirement that they use AI for every task, no matter how small, and work through LLMs first instead of writing code feels like an idea that comes from non-developers looking to make a thought leadership splash. Everyone I know who leverages LLMs uses them as an assistant where appropriate, but trying to go full vibe-code mode where you act through the AI isn't a secret route to more productivity.
This should tell you everything you need to know. It’s not about AI, it’s about using AI to use as an excuse to do what most corporations were already doing: extract more work out of employees without getting them a pay raise, and if they can’t provide that, get rid of them.
What's the point of tools if they don't make people work better?
You just covered exactly that.
Duolingo are not the only ones, I am aware of a project where the whole translation team for internal trainings was replaced by AI automatic translation of training materials.
Any developer that celebrates AI vibe coding, is going to get some bad vides in the coming years.
Wages have lost their relationship with productivity for some time, and while the gains will be privatized, the costs (energy grid, Data center land use, and the carbon emissions) will be born by us all.
So don't think my hostile opinion towards developers that are involved in using AI to abuse their fellow humans is going to be remotely rare in the future. All the necessary preconditions are already there and the only reason it hasn't been noticed yet is because the people that fucked up are still able to get jobs at companies riding AI funding to do AI work.
Look what happened to a lot of the crypto bros. Now multiply that by 10 and the amount of nasty shit they were doing to other workers by 10 and I don't think people are going to take it as lightly as a lot of the crypto bros got off, which was usually just a black mark and severe down leveling when they came back to work at actual companies.
When I ask my collueges than you have the few enthusiasts and then the rest.
The announcement sounds like 'start learning to use These tools' not vibecoding
I hated co-pilot, it kept guessing and then I ended up spending more time reading and altering.. It took me out of the zone.
It was written with 10% higher expectations, 10% higher bar (it was raised in a transformative way) and 10% fewer sloppy thoughts, by a rockstar 100x commenter who uses enterprise AI to blast their KPIs through the pipeline each day.
Sometimes they even circle back to blast their pipeline several times in the same day, all thanks to AI.
Venture capital accepted by Venmo and Apple Cash. Past performance is not indicative of financial statements. GAAP statements may be disrupted and revolutionized at the sole discretion of the company. T’s and C’s apply.
Have wonderful AI day! :)
In fact, to ensure that my comments are 100% more legible and 500% more useful to the end comment consumer, I will leverage AI in all comment responses here on out for at least 10% of my words! This change will ensure excellence and improve results, and most of all, will not in any way lower the high standard of comments you've come to expect from me.
My new Comment Advisory Panel is fully ready to engage with any and all commentors who want advice on how to use AI to make even better comments.
Or trying to write an earnings call headline. The more you mention how much you use AI, the higher your share price climbs!
It's so fucking lousy with gems and upsells and quests and "try the AI call, you get one for free and then you can upgrade to a more expensive subscription!". The gamification of everything in Duolingo is so bad. And, as you say: I feel like I'm wasn't learning any of the fundamentals, can't you just tell me how to conjugate this verb and then testing me instead of making a dumb guessing game out of it?
Compare it to (say) YouTube Premium: I know a lot of people hate it, and that it's an expensive subscription, but I honestly think it's pretty fair. All the commercialization and upsell go away entirely, you get a very clean experience for something that I use more than any other streaming service. Don't mind paying for it at all. Not Duolingo though, I cancelled it after a week and a half.
They would be seen as a stuffy edtech companion app if they did the unsexy work of teaching. Gamification = user growth and engagement = investor money.
For me, Duolingo(uber for Anki flash cards of preliminary words for a 3 day tourist in a new country) was always an odd product. It is very popular among people because they can immediately learn how to say hi/hello, thanks, please in new language but after that, it is akin to learning to swim by reading different tips and tricks, without actually ever touching water or doing the act.
Duolingo was quite an innovative platform years ago, it really led me to keep learning, but now all that's gone. I managed to save every single sentence in the last language I was learning, which I can now import into Anki, so the ad bloat won't be a problem anymore.
They can push their greediness far and beyond, until the extent they find suitable, I won't be part of that.
Nothing about Duolingo gives the impression they actually want you to learn the language. It presents itself as an easy way to start, but if you are more than a single undergrad class into the language and have used any outside resources, it's an obvious waste of time.
Everything on the platform is just a slower form of the most basic note cards. Anki does everything the platform does faster. Anki isn't suitable for all task but Duolingo takes the basic note card and makes you learn at a slower pace.
If finding proper Anki decks for languages wasn't such a massive pain in the ass (Along with navigating the weird 30€ mobile apps for it, are they official, are they not? Can a free alternative do the same?), people would use Anki a lot more.
With Duolingo you can just install, launch, pick a language and get going.
I seriously tried making a deck for my kid for math and holy shit it was a chore and a half.
The same with trying to find a deck for a specific language, no luck there.
For iOS there are about a dozen “Anki” apps? One for 30€, the rest have some kind of in-app purchases.
I’ll rather pay for Duolingo
The SRS approach with cards content made by yourself is more time consuming but works better long term due to the context from which you created the cards, but it takes some time to understand the process correctly (not only the software interface, but how it interfaces with your own memory).
There are also other SRS platforms other than Anki, but Anki gets all the floor space because of its sheer number of users (popularity). Some may be better.
> AnkiMobile is the official iOS app and all purchases help fund Anki's development.[1]
You could always use the web UI in your phone browser, Ankiweb[2], which is very kindly hosted for free by the developers.
You could also write your cards as CSV, HTML, or really any format you want and import them, if the interface isn’t to your liking. Shoot, you can even use an Emacs package[3] if you want to.
[1] https://apps.ankiweb.net/ [2] https://ankiweb.net/ [3] https://github.com/eyeinsky/org-anki
One is made by AnkiApp Inc. (28€-99€ IAP subscription)
One by Ankitects Pty Ltd (30€)
Which one is the official one?
The problem isn't the format of the source, I COULD write it in windows .ini -files. The point is that I want to learn the language, not spend time writing the book about the language first. How do I know what words to add to the deck? Should I add different inflections? How about pronouns, does the language use gendered pronouns? What's the best way to study them, are there rules for it?
I'm willing to pay money for a properly researched and made Anki deck for a language rather than spend time building it word by word.
The best way to study is to make a basic card and just start doing it immediately. A lot of this is individual so using other people's tips isn't going to work (although if you really wanted to do this, you would have very quickly learned by now that there's a whole industry of blog posts and youtube videos and Opinions on how to optimize your deck if you really need that, and there's a whole industry trying to sell you on things to buy to Optimize your Learning Experience). The thing is, overthinking isn't going to work here.
Just make a 2 field note, like the default, and start adding words and doing reviews. You'll very quickly find out what information you find yourself wanting to remember when you're sat there writing sentences in the language, you'll also find out what you find interesting to learn that helps you learn. After the point that you are actually learning, it's really easy to add fields to a note or to switch note types.
Not only are there literally a ton of docs that come with Anki that go over the best way to deal with Anki for learning, but making the deck as you go is ideal for your situation because you're building memory. Inputting sentences and words or whatever you find meaningful to remember into the deck is also building that memory. Spending hours reading blog posts and figuring out which service to pay for, uh, is the opposite of learning.
This might sound harsh, but it literally comes down to "you have to walk to learn how to walk, you have to pedal to learn how to ride a bike". Spending hours or weeks or months deciding if you want to do training wheels or not, what height you want the bike seat, route planning, finding the best bike and the best seat — none of that is riding the bike, all of it is based on preferences that you won't have until you've ridden your first bike, and every inch of it is overthinking and procrastination. Nobody can tell you, beforehand, what the best way to ride a bike is, and that information is meaningless to you until you are physically riding the bike. Nobody can tell you what riding a bike is like and there is no way to learn outside of you physically sitting down on the bike, and pushing off, and pedalling — which is something you can do with a 20$ bike that you found at goodwill. You can, however, find yourself wasting hours or weeks or months thinking about riding a bike and never doing it.
I'd rather just buy a pre-made bike from a store.
Right, but if you were going to do that... you would have done it already, if you catch my drift. Like I said, there's no shortage of blog posts, youtube videos, literal documentation on the Anki website, pre-made decks, etc. available. It's a bit like trying to optimize before you've even started writing code — you don't know what the hot paths are yet, and nobody can tell you without a profiler. You don't know what your preferences are in learning are yet, and nobody can tell you until you try. When you have a solid idea of what you prefer, then people can start giving you recommendations :)
Inputting words is part of using Anki, and a big part of solidifying your knowledge. Why should you listen? Well, even though I haven't reviewed mandarin in literally a decade now, and I've had essentially no IRL use for it, there's still words and phrases I can recall at a moment's notice. The software is unbelievably powerful.
As others have mentioned, there are premade decks but most of the practice of using Anki comes down to preference, so the premade decks may not fit your style of learning at all.
Duolingo is incredibly limited. If you are serious about learning a language, you should look elsewhere. Duolingo is only a good investment of money if you want to pay for a video game with minute levels of learning in the background. It's sort of like playing Sid Meirs Civilization to learn history.
https://ankiweb.net/shared/decks
The difficulty, obviously, is in finding out which decks are good and which aren't. But that's a discoverability problem that is hard to avoid since people are just going to have different requirements and preferences. You can try going by the ratings or check language learning subreddits for recommendations, but ultimately it's up to you to try out decks and see if they work for you. If you just use Duolingo instead, you're arguably not using a better deck, just one with better marketing.
Once you're hooked in, they gamify everything to make you feel like you're doing well. There's the leaderboards, the daily streaks, the ba-ding every time you rearrange 7 word tiles into the only possible sentence that makes any sense.
Go and look at the Duolingo subreddit. When I looked a few years ago, it's just people showing off their 3000+ day streaks.
Anki is a better tool, but its not as polished. Duolingo hooks people with all the little gamified bits, the animations, and it's self-fulfilling popularity loop. Unfortunately being polished is genuinely an incredibly important feature for the majority of people.
These companies tried to quantify the productivity impact of work from home, so it's utterly bewildering to me that they would push these tool-use mandates without actually quantifying the impact LLM tools have on productivity. If it were just 'getting familiar' with AI tools to help define an AI-driven product mindset, I'd expect these CEOs to have more than a naive perception of the tools and their limitations.
I honestly wonder where these mandates started--part of me feels like this is the nascent stage of a VC panic that their AI investment strategy might not work out.
In any knowledge or specialised work, operating a tool faster does not give great results, rather raises the risk of error and quality decline.
Did duolingo once face existential threat because they failed to produce specific feature sometime? Did one of their feature suffer and cause user loss because it took more time for an engineer to write the actual code?
Additionally, beyond formatting and obvious logical errors, every new code should in theory need some human review, which means more automated code means longer review. Assuming code is now produce at 2x, it also balances out that review will now take 2x. Additionally review is much more mentally taxing than putting out one’s thought into code, so risk of bugs and security holes also increases in the long term.
While Software devs cost money, the job involves thinking, and that means it can’t be compared to factory work where someone is standing at next step to simply drill a screw in spotX and the next person will simply put on a cover. Despite years of effort, the attempt to make the process mirror factory floor(did anyone notice the open floor parallel to factory floors?) it failed.
While many hate to see it this way, just like a surgeon will take his time to perform a brain/heart/tumor surgery, a SWE will do thinking, planning, coding and reviews. Giving a surgeon an autonomous bot that can spread the incision area faster or perform the incision faster does not mean productivity gain, it just means the doctor still needs to plan where to make the incision, how much to spread, what to chop off, what to avoid.
The way I have seen this “measured” is by asking (demanding) people to pull some “time saved” number out of their ass. That number is then taken as fact, without question. So the measurable productivity gains are all based on coerced people making things up, omitting instances of the LLM tools slowing things down. It’s a house of cards, and it’s going to fall after a few more months of empty promises without results. “You claimed 300% more productivity, but delivered the same amount of work, and took on this massive AI bill. What the fuck are you doing?”
LLMs are cool, but they aren’t magic. This shit is exhausting, just skip to the part where you fire a bunch a software engineers, because that’s clearly what you actually want to do :/
It’s hard to fathom why this would be the case. Isn’t creating learning materials mostly an upfront task? Obviously you want to update your materials and fix mistakes. And perhaps you sometimes have a new idea that necessitates an update or rewriting things. But creating lesson materials honestly seems like it would scale very well.
I think about reading language learning blogs and forums and people would sometimes recommend favorite resources that were from decades ago. I’m sure something was lost in them not being totally up to date but honestly, persistence is way more important than currency, let alone “having a massive amount of content.”
Encrappification at play.
For a lot of these kind of apps, quantity is the quality in itself.
A policy change like this is designed to have a thin veil of innovation (especially to an unsavy board) but if you read between the lines this is some executive’s wild idea to shake things up because they are completely and totally desperate and really don’t know what to do.
But that is the final stage - AI producing digital content, AI consuming digital content, AI writing code, AI reviewing this code, AI arguing with AI on social media.
No human needed, only numbers growing.
LibreLingo – FOSS Alternative to Duolingo - https://news.ycombinator.com/item?id=43829035 - April 2025 (290 comments)
There is a good possibility that in 5 years people have moved on from Duolingo because new ai tutoring apps that can perfectly tailor content to your level and offer unlimited speaking practice sessions, may surpass Duolingo’s “old-fashioned” offering—-the same way Duolingo jumped ahead of websites, which jumped ahead of CDs and cassette tapes, which jumped ahead of books.
They obviously had to do this. It is already over for Duolingo. All we really need is some kind of Anki deck generator with audio and that plus the language model to have conversations with, it is over. They will look like a childish relic of the past.
Ironically Duolingo removed all of that in the simplification and gameification recently.
> So, the largest part of the market was not being addressed because there was no great way to make money from them. Most people who wanted to learn a language couldn't really afford the best ways of doing it. We wanted to have a way to teach people languages for free. But not just free.
> We wanted to have the best quality of language education, and offer it for free.
Modern Duolingo feels like neither half of that sentence.
Number of people I expect to meet in the future that used "AI first" Duolingo that successfully became fluent in a new language: 0
They don't even really have a functional product to begin with. Meaning that it can take the average person and help them competently speak a new language in a reasonable time frame. Vibe coding I guess can't make it any worse....
I have now been speaking spanish for 9 years and have no use for duolingo when it comes to spanish, but I always recommend it as a resource to level up when you are a beginner.
I'm fluent in French and immigrated here about a decade ago, and I wouldn't have done that if not for Duolingo. It didn't get me anywhere close to fluent itself (Assimil was the single best resource, but no one resource can you get you to fluent), but it got me started and it got me committed. For that, I'm grateful.
I was still in the middle of Spanish course when I realize I can sorta kinda watch and understand some shows, so I watched. (The watching itself then made me progress mucj further, but it would not happen without duolingo).
I still use Duolingo almost daily to have some continuous language exposure, for which I still find it useful (especially as the gamification helps with staying engaged). It has its limitations but it does help me. Just to give a bit of a counterpoint; I find your statement a bit overly broad.
Of course it is possible to learn a language using Duolingo, just like it is possible to get dates on Tinder, but it's just not a good method. If you're new to learning foreign languages, you'd be better off signing up for a course (but that costs time and money), and if this is your n-th foreign language, then you'd rather get a book and some boring flashcard app.
I booked the the single remaining one last year and shortly before the start they announced they will not run it anymore and instead do online classes only. Apparently the rent is too high and it just isn't viable anymore.
It sucks all around :/
Not stuff that's actually useful.
But it still builds vocabulary and is better than nothing for the price.
Just by knowing a few key phrases in the main categories and the probable answers to those in a language will help you get by a lot better than "I like to ride my bicycle in the rain on the weekends" :D
Functional fluency over grammatical competence.
plants for week walk My every uncle takes a his
To be honest though, the main thing that puts me off isn't the teaching quality (which is basic/so-so) but the plethora of weird patterns to keep you hooked. I don't buy the "we want you to succeed" justification. Streaks, streak freezes, begging notifications - anything to keep you looking at ads I guess.
The fluency complaint is completely nonsensical. There is no in person class that would make you fluent, there is no textbook that would make you fluent.
It is possible to criticize Duolingo, but the fluency claim kind of show you don't know what you talk about.
But they say they're "the world's best way to learn a language," right there on their homepage: https://www.duolingo.com.
So either no one has ever successfully become fluent in a foreign language (because not even the best tool can be used to successfully accomplish the task), or the tech industry is full of liars and its claims cannot be trusted.
I happen to have a Duolingo account, and was once a customer. I will send an angry message to their support team of AI robbots and hope it gets carried up to the top.
One of the most valuable determinants for learning a new language is regular practice. Answering 30 easy exercises correctly will do more for your language skill than 10 hard exercises of which you only answer 5 correctly.
And easy questions have the added benefit of being less tedious and convincing more people to stick to the app.
This contradicts the whole thesis statement of their company -- that it is worth time and effort to collaborate and learn with others, even when a machine can make it easier.
Translate the following into English:
jvgug5 g54g 4g g45g ! g54 43 43r pgd0f
Here are the words you need to drag and drop into order:
fish to day cat eat My every eat likes
Do 10 of these every day and you'll be fluent when you hit a 1000+ day streak. Do a lot of these to do more of them than other people on the leaderboard to see your name next to "1." with an icon of a trophy.
500 days of learning various bits & pieces and not being able to have a simple conversation - but I could probably say "There is a monkey in his backpack" if pressed hard!
I used to hate learning from actual textbooks as the conversations felt "dumb" or "forced" but that dumbing down as at least justified by having to progress from zero. Duolingo doesn't feel "plain dumb" but "weird dumb".
So yeah, if they replace their contractors (who must've used the cheapest models) with O3 or o4-mini-high or whatever, it should be an improvement!
It just shows they have no idea what to do, so just squeeze "AI" in... there are too many SaaS out there anyway.
Also "AI first" is BS, until AI has a 100% accuracy it is only useful as long as there are still competent people around that are able to understand what the AI does. A level of competence that gets harder and harder to get in a world where AI assistants allow you to get by by just pressing enter and producing poor quality slop.
Companies and management want to _replace_ human labour because just they don't understand that AI works best _alongside_ people. This doesn't surprise me; one of the worst problems in IT right now is that IT is both pervasive and extraordinarily sector-specific. Capital is in the hands of people that not only don't understand how IT and computers work in detail, but don't even understand how little they understand in the first place
Duolingo’s approach, explicitly tying headcount to proof-of-automation limits, baking AI usage into performance reviews, and prioritizing AI-first systems over retrofitting old workflows, is a glimpse at how "AI-first" won’t just mean using LLMs as a tool, but rebuilding the entire operational model around them.
That said, it's a double-edged sword. Contract workers were crucial to Duolingo’s early scalability. Shifting to AI removes human bottlenecks, but also human nuance — and teaching language is deeply nuanced. It’ll be fascinating (and maybe a little uncomfortable) to see if mass AI content keeps Duolingo's educational quality high as they chase faster scaling.
AI-first might win on cost and speed. But will it still win on outcomes?
Was duolingo ever known for high educational quality? To me duolingo's main pitch was a way to gamify language learning. Of course it became a victim of its own success as soon as you could "pay to win".
Because they focus so much on beginning learners for whom nuance isn't important, this change doesn't seem like it'll hurt them.
Duolingo is widely regarded as more of a game than a high-quality learning experience. People obvious learn something from it, but it's a running joke almost everywhere on social media that people can be 100s of days into their Duolingo streak and still not learn much.
Getting people off of Duolingo and onto less gamified, more rigorous language learning courses is a common theme in the language learning world.
I haven’t used Duolingo in over a decade, but recently I’ve become interested in learning conversational Spanish.
Which is not a terrible strategy. Most people learning languages are doing it for fun or a new years resolution or whatever. If you're serious about learning a language for real (ie you've moved country) then of course you're gonna go to a more serious platform.
It will be a flop. Either it won't get implemented like the C-levels dreamed in the first place and will remain policy on paper only or it will be rolled back quietly once reality hits.
"AI-first" is the "blockchain" of 2025.
AI is increasing our productivity just as the loom did back in the early 1800s. So are HN members now the luddites?
We're both the developers and the destroyers?
Why don't we all just go back to coding on punch cards if we're concerned improved productivity will take our jobs?
We need to look at what we're doing, and what we will be doing in the next 10 or 20 years.
Do we know what that will be? No. But should we get out the pitch forks when a company says "we're going to do more with less people because new technology allows us to do that"?
Also in most software dev task, writing code is rarely if ever the bottleneck, it is the thinking, endless scrum bullcrap, management alignment, architects being too obtuse to not sign off on simpler solutions to protect their prestige, not filling empty roles, pushing teams to extreme and then suddenly deciding to throw away and start something brand new while asking why can’t the thrown away solution be adapted for the new thing or simply consulting/contracting firms intentionally prolonging the project to milk maximum billable hours.
I have actually read the entire screenshot in that tweet and to me it read 50% as non technical executive pitched by some AI vendor about how cool their tools are and 50% of clueless people finding non existent nails for their new hammer.
AI has benefits, it may give productivity gains, specially if you are sweat shops churning out customization of same ERP or wordpress theme. Just repeating productivity gain, coding landscape will change, life will be different 500 times daily because the seller of LLM(calling them the AI wholesale is deceptive, they are subset) wants you to do that, does not magically make those true. The whole trend is sickening and it needs to die.
Replacing SWE is fine, but if SWE can be replaced, the justification can be used to replace crucial things like doctors, lawyers, auditors and lots of other professionals which will kill us all.
A more interesting question is: why are we doing this? Why are people learning foreign languages?
It can be:
- To interact with other human being -> so just find a language exchange
- Because we are bored -> just find another human being
- To find a job / immigrate in an other country -> but it seems AI will replace this job you are after anyway
It is a bit the same with cars. For a long time I wished we had self driving cars because I was commuting 4, 5, 6 hours a day to earn more.
But what would I do when I will have all that money?
Probably enjoy driving an old analog car or better ride my horse everyday.
Also anybody here eating poison with the dream of growing your own food?
So always look at the big picture and you will see DuoLingo is irrelevant anyway. It is another robotic hamster wheel like Instagram and TikTok.