Two that didn’t work
Not everything I build takes off. These two didn’t, and I keep them here on purpose: the projects that worked look the way they do because of what these two got wrong.
- ROLE
- Everything, both times. Code with AI coding agents.
- PROJECTS
- productcant.com (2026), 100dollar.jobs (2025–2026)
- STACK
- Node scripts + GitHub Actions; Next.js, Supabase, n8n
- STATUS
- One runs on its own, one is a domain for sale
Product Hunt and productcant’s own wall, August 2026. Listings and subscribers: the 100dollar.jobs database, August 2026.
01 PRODUCTCANT.COM
It works, and nobody comes
productcant.com is a public record of the Product Hunt launches nobody came to. Every day a script reads the new launches, sorts them by upvotes and files the quiet ones on a wall. Most launches finish on two upvotes; a handful take almost all the attention. I bought the domain on 14 July and it went live six days later.
It started as a site that mocked big promises. When I finally measured my own filter, 66 of the 70 launches on the wall had qualified for nothing more than the word “AI” or “platform” in their tagline. It was laughing at a tool for restoring family photos. So I turned it around: the site now gives those launches attention instead of taking it away, and the joke lives only in a weekly ballot aimed at outsized claims, never at people.
On 4 August I launched it on Product Hunt, mostly for the irony. It finished on four upvotes, and the next daily build filed it on its own wall. There is no exception in the code, and I didn’t add one.
What went wrong wasn’t the code. It was focus. I spent my time on the type scale, the ballots and the data pipeline, and almost none on getting it in front of people: the Show HN post is written and has never been posted. It still runs every day on its own, which is the most honest thing I can say about it.
02 100DOLLAR.JOBS
A domain looking for a product
This one started backwards: with a domain I liked, and no idea what to put on it. It was also my first project built with AI coding agents, first in Antigravity, then in the terminal. Everything I now do every day, I first tried here.
Three versions, one after another, then a sign:
- Early 2026A job aggregatorFreelance gigs scraped every two hours, about 21,000 in all
- May 2026The path to $100/hourRates for 20 skills, interviews, an AI rate audit
- Jul 2026Human vs AI agentThe same task priced for a person and for an agent
- Aug 2026For saleNever launched, sold as it is
Each version was finished. None was ever really launched. The posts for X and Hacker News were written and never sent, and in three weeks the last version got zero real subscribers. In August I decided to sell it as it is. Today the domain is listed for $990.
03 WHAT THEY HAVE IN COMMON
Both stopped where the building ended
Neither failed on the tech. Both worked. Both stopped at the same point: the build was done, and getting it to people was a separate job I kept putting off.
So in the projects that came after, distribution is part of the product from day one. The F1 app has a social machine that writes its own posts and I only approve them. BiałkoRadar sends people a push when a price drops. If nobody is going to hear about it, it isn’t finished.
04 SEE THEM