I built 128 things with AI. 127 are dead. So I started shipping into real OSS instead.

A Japanese developer built 128 AI projects in four months, but 127 failed due to poor distribution. Instead of building another app, they started contributing to open-source tools by localizing them into Japanese, fixing issues like half-translated UIs and IME bugs. They created a live counter tracking merged pull requests to provide verifiable proof of their work.

i built 128 things with AI in four months. then i made an AI dissect every one of them, and the verdict was uncomfortable: good at building, beginner at distribution, and 127 of them are already dead. so i stopped building app 129. instead of shipping into the void again, i started shipping into tools people already use. in one day i opened pull requests localizing real open-source dev tools into Japanese — Hoppscotch, Jan, Betterlytics. here's why it matters. i'm Japanese, Tokyo-based, and a lot of great tools quietly fall apart for Japanese users: half-translated UIs, the Enter key that fires mid-IME-conversion and submits your half-typed text, a /ja locale with nothing routed to it. small, annoying, invisible-in-analytics stuff. exactly the stuff i can actually fix. then i put a counter online that reads straight from GitHub's API: https://greymoth-jp.github.io/proof-dashboard https://greymoth-jp.github.io/proof-dashboard i can't fake that number. when a maintainer merges my code, it ticks up on its own. that's the whole point — in a feed drowning in AI-generated "i shipped X" posts, the last thing that's genuinely hard to fake is code that passed someone else's review into production. i'm going to keep doing this, one PR at a time, in the open. so two questions, and i actually want answers: