Meet Daddy Dativ – the German tutor I vibe coded in 3 days, rather than 30
What building an app with AI made me think about the future
Stretch goal reached! Last week, I announced I’d be using AI to build my own German tutor. I estimated it would take me a month, and maybe I’d be able to hit some stretch goals. In the end, it took me just three days to start hitting my stretch goals. I’m blown away.
In this article:
First, I’ll run you through the app I built and explain how I developed it.
Then, I’ll dive into the implications for software, design, and our interaction with the digital world.
Building Daddy Dativ
After sending out last week’s newsletter, I learned about a vibe coding app called Lovable through Rex Woodbury’s newsletter, Digital Native. I decided to give the 5 free prompts it offers a go. I registered, and typed into the textbox on the front page:
“Build me an app that teaches me how to finish words in German in different cases”
It immediately spun up something decent, and within half an hour and a few nudges, I had an app that was 80% done. I was shocked because I actually did this during my German class, which sometimes gets a bit too easy due to similarities between Dutch and German grammar. A distracted me, with AI, could build in about 30 minutes what a focused me would have struggled to build in weeks a few years ago.
A distracted me, with AI, could build in about 30 minutes what a focused me would have struggled to build in weeks a few years ago.
I kept toiling away at it for a few days whenever I had time, and in the end, I have exactly the app I needed. What’s best is that I can add functionality and content whenever I want.
Some of Daddy Dativ’s features:
Multiple choice questions to practice cases for adjectives, pronouns, articles, and question words.
Lots of grammar-themed dad jokes, including a joke of the day on the front page and a new dad joke every time you take the quiz.
Tips for better grammar learning.
An overview of your answers after each round.
Responsive mobile-friendly design. This was a stretch goal, as I had initially focused on desktop-only solutions.
A top scorer feature that lets you swap out the YouTube video on the front page whenever you answer at least 90% correctly during a session. I know this is sensitive to trolling, which is by design: it functions as an incentive for people to test the app.
Nearly all the content was created through Lovable (which uses Claude 4), though I also added in some questions I generated myself with Claude Sonnet 4 and ChatGPT-4.
I had to triple-check all the exercises, as I found some minor mistakes. I also dropped “fill in the gap” type exercises from this version, as those were more prone to LLMs making errors. I will reintroduce them eventually, as the feature was working properly, but I’ll need some extra time to verify that content and for now, the multiple-choice version is providing enough of a challenge. No need to bloat the app with features when the user’s need is met.
Overall, I spent 20 euros and 20 hours to create my ideal language-learning app. The number of hours would have been lower had I decided to keep this app just to myself (no polishing, no cookie banner, no YouTube feature, no terms & privacy pages, etc.). That’s pretty great.
Some of my main takeaways:
The last 20% will take 80% of your time. ‘Vibe debugging’ is painful, as it can be challenging to follow along with what’s going wrong. This is where understanding some basics of software development really helps.
Converse with your AI to verify that you have the same understanding of what needs to be changed. In some cases, I prompted it for changes, and it altered parts of the app that it shouldn’t have been touching at all. From then on out, I asked it to present me a list of all changes each time before I put it to work. You can always roll back, but it’s a bad way to spend your subscription credits ($25 gets you 100 credits, which is basically fuel burned every time you interact with the AI, with more being burned for more complex interactions).
Have the AI create a little dashboard, so you can test out the various states of the app. For example, efficiently testing the top scorer feature gets complicated if I have to do a new test with a 90% score every time.
Build something that you see yourself using over and over. This will make it fun to build. This will keep you motivated through the drudgery of the last 20%.
If you end up building something for yourself, feel free to share with our Signal group. I’ll continue tinkering this month and will share updates on Bluesky and Substack Notes. If you end up trying Lovable, we both get 10 extra free credits if you use this link. I’m curious to hear about your experiences in the comments or hit reply to this email.
And for the German-language learners, please let me know what you think about Daddy Dativ.
Implications for the future
Considering the future, it would be a mistake to assume the current status quo and extrapolate from it. Five years ago, vibe coding didn’t really exist. There were no-code app builders, but they were not comparable to what’s possible now. Two years ago, this existed, but was much more limited. This type of functionality will continue to improve, and as it does, the context of how we interact with the internet will also change.
So, really, this isn’t about implications for tech enthusiasts only, but really about a shift in the everyday internet for everyone.
As I started my Daddy Dativ project, Techdirt founder Mike Masnick wrote about his own vibe coding project. He built a task management app that caters explicitly to his own needs, and as someone who has written about tech through a critical lens for decades, he has some great takeaways:
Personal agency: The key implication Masnick highlights is that people can now escape the tyranny of ‘one-size-fits-none’ software. Rather than adapting your workflow to align with some product manager's vision of the ‘average user’, you can create tools that align precisely with how your mind works. This signifies a shift from renting someone else's productivity philosophy to owning your digital workspace.
Democratisation: Technical barriers to building custom solutions have largely collapsed. Like me, Masnick developed a working productivity app in three days, using nothing but conversational prompts. This means non-programmers can solve their specific problems without waiting for companies to prioritise their use cases or learning to code.
Freedom from subscriptions: Instead of paying recurring fees for tools that only partially meet your needs, you can create precisely what you desire at minimal cost. Masnick’s app, like mine, incurs almost no operational expenses and cannot be "enshittified" by growth-focused product decisions or abruptly discontinued.
True and rapid customisation: When you need a new feature, you just prompt for it. You can add features when you need them and get immediate solutions, rather than hoping some company might eventually build similar functionality and then potentially upsell you on it.
Solving the personalisation gap: Many mass-market tools get users 60-70% of the way to what they actually need. Vibe coding lets you build that final 30-40% so you can build exactly what works for you.
Another vibe code project that hit my timeline was a CEO simulator built by Adam Arrigo, one of the founders of Wave, a 3D immersive entertainment platform. It’s not simply a CEO simulator — it specifically simulates some of the decisions and trade-offs Arrigo faces as Wave’s CEO. It’s a simple game he built with Claude over a weekend that can give investors, team members, and partners better insights into how a company like this operates.
For me, this is an excellent example of small context creativity, which I wrote about a few weeks ago. A reminder:
“Due to the proliferation of easy-to-use image editing software, we got internet meme culture, which has gotten incredibly context-specific. We now have niche memes for our local workplaces, yoga studios, techno clubs, family group chats, newsletters, etc.”
The learning curve and effort required to build my own app years ago would have prevented me from doing it. Now I can build one every week. This means it won’t just be developers who can create gimmicky software as memes: everyone will be able to express themselves through software. It’s software as language.
But it also means it’s viable to create complex apps that just have one user — you.
Apps — as we know them — are dead
People increasingly turn to ChatGPT for tasks they would previously have used Google for, such as finding recipes, researching products, and looking up information. A big reason for this is the so-called ‘enshifttification’ of the web. People want to get to utility as fast as possible, and they don’t want to see ads or read long personal stories on recipe pages, etc. So, ChatGPT is a good way to bypass that: it allows people to personalise their web experience and cut out everything they don’t need.
ChatGPT’s personalisation of web experiences is the continuation of a trend we already see on content-centric platforms. Consider the recommendation algorithms on music streaming services or video-on-demand platforms that ensure the app’s content is relevant to you, in exchange for your prolonged engagement and subscription revenue. Or social media apps’ algorithms that keep people tied into their custom feeds so that they see more ads. So, where to next?
LLMs are going to get better and faster at spinning up software.
LLMs will better understand who is using them and their preferences.
People will increasingly rely on LLM apps, like ChatGPT, to interface with anything digital.
That’s why I think of these LLMs as operating systems. We’re currently close to a point where anyone can spawn entire apps with a few prompts. Today, I asked Claude if we could build something that lets me check my history for interesting links I could add to this newsletter every week, and it just built me something. I’m not a paid user. I didn’t explicitly tell it to. I just asked. It just provided me with an interface to do what I’d asked about.
This is not the future of the internet. It is here already. It’s just not how most people use the internet yet.
Today: widespread ChatGPT use. People spin up mini-apps in the form of prompts to solve basic, everyday problems. Answers are usually returned in a conversational format.
Tomorrow: people will engage with AI to let LLMs generate custom user interfaces, tailored to their or their team’s specific needs.
We’ve been here before. Back in the late 2000s, when Facebook launched its developer platform, the tech press speculated a lot about the emergence of Web Operating Systems, of which Google’s suite of Drive, Calendar, and Gmail is currently probably the best surviving example. The path towards web-based operating systems diverged due to the rise of smartphones, iOS’s App Store, and Android / Google’s Play Store.
We’re again at such a point, though it’s hard to predict how things will fall into place this time (consider the convergence with smart devices and wearables, for example). For now, the best way to get a taste of the future is to explore. Build something for yourself. Host it on your own server.
Masnick ends his article with a paragraph that starts with the following:
“So, for all the folks who are concerned about centralized control and losing agency over their digital lives: there are now more options beyond just yelling at tech companies or hoping for better government regulation.”
He’s totally right here. Make your own software. Make apps to help your friends.
Small software is here.
For our regular readers: unfortunately, no extras for your ears and eyes today.
See you next week. <3
What’s your vibe coding go to at the moment? In Claude? I’ve been playing with loveable, Vervel V0, and Ohara.ai