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.
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.
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.
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.
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.”
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.
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.
> 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....
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.
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.
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.
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?
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"?