A lot of indiepreneurs and founders tout the importance of being anti-fragile. I did quite the opposite. I turned a great opportunity into a lagging product. In a way, I flipped the order.

This is my story of how I kept a great potential collecting dust for two years.

Circa 2023...

When the GPT API was first released, I was one of the quick to create an agentic news notification system with GPT, first starting with Hacker News. I had a couple of reasonings and philosophies while building it.

• News should come to me, not vice versa.News should come to me, not vice versa.
• Tangible AlgorithmTangible Algorithm
• i18n-nativei18n-native
• Open SourceOpen Source

This is why I first built what's now known as Heimdall. It was a crude agent that ran a cron job on the most popular HN stories and translated them into 30 different languages.30 different languages.This was still an infancy era of agentic web access, so I built lots of layers by myself, like providing recursive feedback to the agent, error handling, agentic web crawling, and simple RAG, albeit my version was very rudimentary.

The reactions were simply amazing.

I made almost no effort in public relations, except for a tiny URL on my MikeRowSoftedMikeRowSofted extension, and people flocked in through pure word of mouth—from Italy to China, not to mention my home country, Korea, and the US.

I have gathered 2K daily users and sent 1M notifications to this day. The open rate was consistently over 80%over 80% every day—almost unheard of in the email business.

People started requesting features.

Soon, people started to ask me for features, from supporting their languages to fine-tuning the translation scripts. But most importantly, they wanted summaries on other topics as well. That's when I started building the second version of this project, which allows natural language search queries. Think of it as Google Alerts but with more in-depth research and briefings. Eventually, I built a small MVP, went to a hackathon, and became one of the winners. I got some AWS credits and a small prize, and things were all smooth sailing.

That went only so far.

The problem was that I was really bad at anything other than front-end. I didn't know how to use databases, servers, or infra stuff. As a university student with primarily front-end experience, things weren't working well.

Most importantly, vibe codingvibe coding did not exist back then. Eventually, I was frustrated and left the product, doing very minimal life support on it. I left the organic 2k users and 80% of open rates daily. As you might've guessed, if I were anything agile or antifragile, I would've already put this out in the market.

Full-time Work and Advances in Abstraction Layers

I started my full-time work in the Medical MLOps field. This was one of a kind, requiring me to be a genuinely full-stack engineer. I eventually led a complete rebuild of our MLOps console with RSC with all the latest bleeding-edge technology, and I was lucky to have my team support me.

I learned a lot more about managing the back side, such as the Backend, Database, Cloud, Redis, and CI/CD. More importantly, I learned how the traditional way of doing the backend is becoming increasingly streamlined with the latest products on platforms like Neon, Upstash, and Vercel. I now understand why those are worth the hype and how they are helpful.

Then I remembered my abandoned brainchild.

I wanted to build it again. Luckily, there are no similar products yet. If I want to poll new info on "New Developments in React periodically," I need to build tools one by one with tools such as n8n or julep. There is no click-click-done SaaS for bespoke AI news alerts. So here's the deal: I want to restart the work on this.

I am going to bootstrap this anyways... But here's how Vercel OSS program can help.

I know what it's like to build satisfying interfaces.

Components deliver intent and experience. That's why Vercel's recent 'gotta catch em all''gotta catch em all' is so interesting. However, due to the lack of proper view transitions and stateful navigations, no component libraries match iOS-level fluidity. That said, during my process, I want to build open-source component libraries for others. This may take months and years, but I want to accelerate my ideation and prototyping process and give great examples back to the community. Here are some components I built, with ergonomics in mind.

I know Next.js. Very well.

I know and use the latest bleeding-edge technology, such as RSC, to build this. While I cannot disclose the source code of the internal web app that I built for my company, it loads 50M data points from archives in the Netherlands to Korea in less than a second. I doubled down on understanding how Next.js caching works and eventually led a projectled a project to rebuild the entire frontend into Next.js 15 App Router for our internal web app.

While a lot of legacy media websites use Next.js, I can promise no single company will know Next.js App Router Cachings and PPRs better than I do. Here's a small demo of directory navigation in the internal web app I built for my company. We had a bunch of nested layers, and we had to navigate them quickly, and eventually, I created this spatial cognitive design.

Not only ergonomic, but also exceptionally fast. These are dummy data (NDA), but prod is on the same speed.

I am i18n-native. More than anyone.

Based on my previous philosophies, I want to keep this available to many people. In the v1 version, most of the costs came from avid users in developing regions like South East Asia or Africa. They loved the product because they could follow Silicon Valley news at near-identical speed as San Franciscans. The reader base is small, but it cost me a good chunk of token money, which made me consider dropping support. However, I remain committed to making this as i18n-native as possible.

So here's the deal.

I will finish building this byMay 6thMay 6th, and let people sign up.

I understand you'd be concerned that I did not build the app after the previous hackathon. But now I know everything better, and more importantly, I have better self-discipline. Therefore, if I do not finish building the app by the end of May 6th, I will refund your money and pass it on to the next candidate or donate it to an OSS software of your choice.

I will ship subpackages this time.

I mentioned that I handmade a lot of the stacks on v1. This included custom scrapers, IP rotators, search engine integrations, summary generators, email handling, and more. If you've been following the Agentic Framework worlds, you'd immediately notice that most of the "layers" have now spun into a whole industry, such as Exa AI, Firecrawl, Resend, Kagi Search API, and so much more. While mine was rudimentary, I built them earlier than anyone. I didn't ship them (which is my fault and my bad.) This time around, while utilizing these open market tools, I have so many ideas that I can add to the stack, such as agentic feedback collectingagentic feedback collecting, agentic A/B testingagentic A/B testing, vector-storage personalization memoryvector-storage personalization memory, agentic war roomagentic war room and so much more. I will also open-source them and make those readily available.

Triple-Down Effect

While economists debate the "Trickle-down" effect, I can ensure that Sunghyun's new Triple-DownTriple-Down economics will definitely work. I already sponsor more open-source projects than many of the companies just cuz I think they're cool. If you sponsor me with this project, I will for sure triple the amount and pass it down to the subdependencies of my journey.

So...

I understand that I am in a unique position, making it hard for Vercel to support me. This is more of my manifesto instead of an appeal. It is a journal and a record for me to discipline myself and put myself back into builder mode. So, I hope you enjoyed reading my anti-agile manifesto.

Sincerely,