Show HN: Search 17 years of "What are you working on?" (stillworkingonit.vercel.app)

1 points by EagleEdge ↗ HN
Inspired by Karpathy's HackerTube, I wondered what was buried in "What are you working on?" thread comments. I wanted to see what changed, and who was still at it years later. So, I built this site to answer some of those questions.

A couple of things you can do: type in your idea and see who already tried it, look up your username and see what you were working on in 2020, see who’s been working on the same thing for a decade, or what founders posted before they had a company.

I built it using Claude Code. Claude Sonnet read all 20,982 top-level replies and pulled out a name, description, topic and how AI was involved. One finding is that AI went from 1% of answers in 2009–11 to 33% this year.

A few limits of the result: labels are model output, and I only spot-checked them. The HN thread only went monthly in 2024, so early years are thin. If I’ve mislabeled your project, or you’d rather not be included, please tell me.

If you’ve been posting here for years, I’d love to hear what happened to those projects.

1 comment

[ 7.1 ms ] story [ 23.6 ms ] thread
Several quality improvements I have made and new insights discovered since the launch of this site:

1. How long projects last. Each project's site and repo is now followed over time (8,843 sites through the Internet Archive, 2,802 repos through GitHub). 51% of sites are still up after five years and 31% after ten. For repos still being updated it is 33% and 15%. 2. Silence is not an ending. Only 12% of projects are mentioned again within six threads. But a project that is still being answered about shows up in only about one thread in five, so not seeing it again says little. 3. Journals. The 1,807 projects answered about in more than one month now read as a dated story: their answers, their HN stories, and 603 short quotes in the authors' own words. 4. Search shows outcomes. A search for an idea now says what became of the projects that match: how many came back, reached 100 points on HN, have a starred repo, or are still going a year on. 5. The thread, live. The current thread is read straight from Hacker News, with each answer shown next to that account's history on the site. 6. Accuracy is published. The labels were measured against 600 answers labeled twice independently: 93% for kind of answer, 80% for topic, 93% for the role of AI.